This document discusses artificial intelligence applications in radiation oncology. It begins with acknowledgements and then outlines several AI applications including radiomics tools for lung cancer screening, tumor response prediction, and predicting aggressive lung adenocarcinoma subtypes. It also discusses using AI for automatic tumor delineation and quantification of delineation variability as well as local tumor morphological changes prediction and metabolic tumor volume changes. The document provides details on methods and results for several of these AI applications in radiation oncology.
Artificial Intelligence in Radiation OncologyWookjin Choi
- The document discusses the use of artificial intelligence and radiomics in radiation oncology. It presents frameworks for radiomics analysis involving image registration, tumor segmentation, feature extraction, and predictive modeling.
- Specific applications discussed include using radiomics for lung cancer screening and prediction of tumor response. Radiomics features combined with machine learning models show improved performance over clinical guidelines for assessing lung nodule malignancy.
- Methods are also presented for quantifying tumor characteristics like spiculation through image analysis and extracting interpretable radiomics features. This can provide semantic information to radiologists for assessment.
Artificial Intelligence in Radiation Oncology.pptxWookjin Choi
The document discusses artificial intelligence applications in radiation oncology, including automatic delineation of organs-at-risk using deep learning models like OARNet. It also discusses radiomics approaches for clinical decision support and outcomes prediction using features extracted from medical images with techniques like spiculation quantification for lung cancer screening.
Treatment verification systems in radiation therapyanju k.v.
This document discusses various techniques used for treatment verification in radiation therapy. It describes electronic portal imaging devices (EPID) which can be used for daily treatment localization and verification through portal images with little additional dose. Cone beam computed tomography (CBCT) is also discussed, which provides volumetric CT images with submillimeter resolution, allowing verification of patient positioning before treatment. Both EPID and CBCT help ensure the correct radiation dose is delivered to the intended target volume.
1. Dr. Sheetal R Kashid presented on the use of IGRT for head and neck cancers and central nervous system tumors at TMH.
2. IGRT uses image guidance to precisely position patients and correct for setup errors, allowing for accurate radiation delivery while minimizing dose to surrounding healthy tissues.
3. At TMH, IGRT is performed using CBCT, EPID, and offline protocols to correct for systematic and random errors in head and neck and neuro-oncology patients.
Stereotactic body radiotherapy (SBRT) delivers high-dose radiation to tumors in a small number of fractions using high precision. For prostate SBRT, the target and organs at risk are contoured on planning CT. A dose of 35-38Gy in 5 fractions is used as primary treatment for low risk prostate cancer. Rigid image guidance and intrafraction monitoring are important to minimize setup errors. ExacTrac X-ray positioning co-registers X-rays with digitally reconstructed radiographs and corrects for rotational and translational deviations, achieving sub-millimeter accuracy. This allows safe dose escalation for prostate SBRT.
A summary of recent innovations in radiation oncology focussing on the priniciples of different techniques and their application. An overview of clinical results has also been given
Artificial Intelligence in Radiation OncologyWookjin Choi
- The document discusses the use of artificial intelligence and radiomics in radiation oncology. It presents frameworks for radiomics analysis involving image registration, tumor segmentation, feature extraction, and predictive modeling.
- Specific applications discussed include using radiomics for lung cancer screening and prediction of tumor response. Radiomics features combined with machine learning models show improved performance over clinical guidelines for assessing lung nodule malignancy.
- Methods are also presented for quantifying tumor characteristics like spiculation through image analysis and extracting interpretable radiomics features. This can provide semantic information to radiologists for assessment.
Artificial Intelligence in Radiation Oncology.pptxWookjin Choi
The document discusses artificial intelligence applications in radiation oncology, including automatic delineation of organs-at-risk using deep learning models like OARNet. It also discusses radiomics approaches for clinical decision support and outcomes prediction using features extracted from medical images with techniques like spiculation quantification for lung cancer screening.
Treatment verification systems in radiation therapyanju k.v.
This document discusses various techniques used for treatment verification in radiation therapy. It describes electronic portal imaging devices (EPID) which can be used for daily treatment localization and verification through portal images with little additional dose. Cone beam computed tomography (CBCT) is also discussed, which provides volumetric CT images with submillimeter resolution, allowing verification of patient positioning before treatment. Both EPID and CBCT help ensure the correct radiation dose is delivered to the intended target volume.
1. Dr. Sheetal R Kashid presented on the use of IGRT for head and neck cancers and central nervous system tumors at TMH.
2. IGRT uses image guidance to precisely position patients and correct for setup errors, allowing for accurate radiation delivery while minimizing dose to surrounding healthy tissues.
3. At TMH, IGRT is performed using CBCT, EPID, and offline protocols to correct for systematic and random errors in head and neck and neuro-oncology patients.
Stereotactic body radiotherapy (SBRT) delivers high-dose radiation to tumors in a small number of fractions using high precision. For prostate SBRT, the target and organs at risk are contoured on planning CT. A dose of 35-38Gy in 5 fractions is used as primary treatment for low risk prostate cancer. Rigid image guidance and intrafraction monitoring are important to minimize setup errors. ExacTrac X-ray positioning co-registers X-rays with digitally reconstructed radiographs and corrects for rotational and translational deviations, achieving sub-millimeter accuracy. This allows safe dose escalation for prostate SBRT.
A summary of recent innovations in radiation oncology focussing on the priniciples of different techniques and their application. An overview of clinical results has also been given
This document discusses the use of stereotactic body radiation therapy (SBRT) for liver tumors. It provides details on common liver tumors including hepatocellular carcinoma and metastases. It describes SBRT as a treatment option for inoperable early stage tumors, as a bridge to transplant, and for intermediate or locally advanced stages. Key factors for patient selection and treatment planning such as tumor size, number and location, as well as liver function are summarized. The document also briefly discusses proton beam therapy and current clinical trials investigating SBRT for liver cancer.
Motion management strategies in radiation therapy aim to account for tumor movement during treatment. Key strategies include gating methods that deliver radiation only during specific respiratory phases, breath hold methods that immobilize tumors during deep inhalation or exhalation, tracking methods that follow tumor motion in real-time and adjust beam targeting accordingly, and encompassing methods that define larger target volumes to cover full respiratory excursion. No single approach is clearly superior, as appropriate management depends on tumor location, motion extent, and available technology. The goal of all motion management is to safely escalate dose to tumors while reducing dose to surrounding healthy tissues.
This document summarizes the process of simulation for radiation therapy treatment planning from CT imaging to treatment verification. It describes how patient positioning is done using lasers during CT scanning and how the CT images are imported into the treatment planning system. It also explains how the treatment planning system localizes CT markers and defines the isocenter in machine coordinates for treatment. Finally, it summarizes the verification process of aligning the patient using digital reconstructed radiographs and portal images to ensure accurate treatment delivery.
TBI is the radiotherapy technique to irradiate whole body before doing stem cell transplant. The main purpose of doing TBIB is to condition the immune system of body so that there will be maximum chance of transplant acceptance.
This document discusses techniques for simulation, planning, and treatment delivery for stereotactic body radiation therapy (SBRT) for liver metastases. It covers important steps including patient preparation, positioning, immobilization, motion management, imaging, and treatment execution. Motion management techniques discussed include abdominal compression, breath holding, gating, and tracking using internal or external surrogates. The importance of accurate simulation and reproducibility is emphasized for precise SBRT delivery.
This document discusses the importance of treatment verification in radiotherapy and outlines the process. It notes that even small errors can have negative consequences so treatment verification is essential to ensure the right dose is delivered to the right area. The key aspects of treatment verification are machine setup, monitor units, patient positioning and imaging by comparing images to references. Errors can be systematic from planning or random from daily variations; various methods are described to reduce errors and ensure treatments are accurately delivered.
Final ICRU 62 ( International commission on radiation units and measurements)DrAyush Garg
The document discusses recommendations from reports by the International Commission on Radiation Units and Measurements (ICRU) for defining volumes used in radiation therapy planning and reporting. ICRU Report 62 provides additional details on volumes such as the internal target volume (ITV) and planning organ at risk volume (PRV), and introduces metrics like the conformity index. It also further classifies organs at risk as serial, parallel or serial-parallel based on their radiosensitivity.
The vmat vs other recent radiotherapy techniquesM'dee Phechudi
VMAT is a new type of intensity-modulated radiation therapy (IMRT) treatment technique that uses the same hardware (i.e. a digital linear accelerator) as used for IMRT or conformal treatment, but delivers the radiotherapy treatment using a rotational or arc geometry rather than several static beams.
This technique uses continuous modulation (i.e. moving the collimator leaves) of the multileaf collimator (MLC) fields, continuous change of the fluence rate (the intensity of the X rays) and gantry rotation speed across a single or multiple 360 degree rotations
This document discusses image-guided radiation therapy (IGRT) and its evolution and applications. It begins by defining IGRT as external beam radiation therapy using imaging prior to each treatment fraction to verify patient positioning. IGRT allows for reduction of safety margins by compensating for set-up errors and organ motion. The document then reviews the history of IGRT from early portal imaging to modern cone-beam CT and other volumetric imaging techniques. It provides examples of IGRT protocols and clinical outcomes for sites such as prostate, lung, liver, and central nervous system tumors.
The document discusses the use of Tomotherapy for radiation treatment planning and delivery. It provides examples of how Tomotherapy allows for:
1) Highly conformal radiation plans that sculpt dose around complex tumor target shapes while minimizing dose to nearby organs.
2) Daily image guidance that enables adjustment of targets to account for changes in patient anatomy and tumor size during treatment.
3) Delivery of simultaneous integrated boosts to multiple tumor sites.
This document discusses 5 reasons why radiology needs artificial intelligence:
1. There is a global shortage of radiologists that is expected to worsen as imaging volumes increase faster than new radiologists enter the field. AI can help improve radiologist productivity.
2. AI can enhance radiologist productivity through functions like smart alerts, automatic image annotation, and faster access to patient information.
3. AI can help improve diagnostic accuracy by reducing human error, tracking lesions over time more quantitatively, and providing automated second opinions.
4. AI may help lower misdiagnosis rates by alerting radiologists to disease indicators and comparing new cases to existing ones to reduce fatigue-related errors.
5. AI could help improve
An overview of imrt, vmat optimization algorithmsRahim Gohar
This document discusses intensity-modulated radiation therapy (IMRT) and volumetric modulated arc therapy (VMAT) optimization algorithms, including direct machine parameter optimization (DVO), leaf motion calculation (LMC), progressive gradient optimization (PGO), and multi-resolution dose calculation (MRDC). It emphasizes that theoretical physicists developing these algorithms have revolutionized the field of radiation oncology by enabling more conformal radiation treatments.
This document provides an overview of image-guided radiation therapy (IGRT) for lung cancer. It discusses the role of IGRT in managing tumor motion through techniques like breath hold methods, free breathing with gating or tracking, and 4D imaging. Segmentation of the tumor and organs at risk on 4D CT scans is covered. Dose fractionation schedules and biological effective dose calculations for hypofractionated stereotactic body radiation therapy are reviewed. Toxicities, outcomes, and challenges of IGRT in lung cancer are also mentioned.
Basic information about Elekta and its familiar with xvi and Iviewgt protocols and there import and defining the Target area clip box registration along with HEXAPOD 6Dof couch & Apex Dmlc setup
This document discusses artificial intelligence and its applications in radiology. It begins with definitions of artificial intelligence and its subsets of machine learning and deep learning. It then discusses how machine learning and deep learning are being used in medical imaging for tasks like cancer diagnosis and detection of findings in images. The document outlines how large amounts of medical image and patient data are being used to train AI models to perform tasks like segmentation and anomaly detection. It provides examples of startups and projects applying AI to problems in radiology. It concludes by discussing views on AI in radiology, noting that AI can increase radiologist efficiency and consistency if integrated into the workflow, rather than replacing radiologists.
This document summarizes the expectations and key learnings from a linear accelerator acceptance, commissioning, and annual QA training that occurred from September to November 2008. The training covered:
1. Fundamental concepts of linear accelerators, beam production, safety features, and the acceptance testing process.
2. Techniques for collecting beam data needed for commissioning, including measurements and data definitions.
3. Procedures for linear accelerator QA and other treatment machine QA on an annual basis.
Key topics included the beamline components that produce photon and electron beams, characteristics of linear accelerator beams, the importance of acceptance testing and commissioning the machine properly, and techniques for annual QA tests.
Three dimensional conformal radiotherapy - 3D-CRT and IMRT - Intensity modula...Abhishek Soni
Conformal radiation therapy techniques like 3D CRT and IMRT aim to concentrate radiation dose in the tumor while sparing surrounding normal tissues. This is achieved through advances in imaging, treatment planning and delivery. 3D CRT uses geometric field shaping with multiple beams while IMRT further modulates beam intensity across each field. Both require contouring of target and organs at risk on imaging along with inverse or forward treatment planning to optimize dose distribution. Conformal techniques allow higher tumor doses with improved normal tissue sparing compared to conventional radiation therapy.
Artificial Intelligence in Radiation OncologyWookjin Choi
The document discusses artificial intelligence applications in radiation oncology. It begins with acknowledgements and then outlines topics including radiomics decision support tools, automatic delineation and variability analysis, and applications like lung cancer screening, tumor response prediction, and aggressive lung adenocarcinoma subtype prediction. Radiomics frameworks and deep learning models are presented. Results show potential for AI to provide quantitative imaging biomarkers and improve outcomes in areas like screening, treatment planning, and response assessment.
This document discusses the use of stereotactic body radiation therapy (SBRT) for liver tumors. It provides details on common liver tumors including hepatocellular carcinoma and metastases. It describes SBRT as a treatment option for inoperable early stage tumors, as a bridge to transplant, and for intermediate or locally advanced stages. Key factors for patient selection and treatment planning such as tumor size, number and location, as well as liver function are summarized. The document also briefly discusses proton beam therapy and current clinical trials investigating SBRT for liver cancer.
Motion management strategies in radiation therapy aim to account for tumor movement during treatment. Key strategies include gating methods that deliver radiation only during specific respiratory phases, breath hold methods that immobilize tumors during deep inhalation or exhalation, tracking methods that follow tumor motion in real-time and adjust beam targeting accordingly, and encompassing methods that define larger target volumes to cover full respiratory excursion. No single approach is clearly superior, as appropriate management depends on tumor location, motion extent, and available technology. The goal of all motion management is to safely escalate dose to tumors while reducing dose to surrounding healthy tissues.
This document summarizes the process of simulation for radiation therapy treatment planning from CT imaging to treatment verification. It describes how patient positioning is done using lasers during CT scanning and how the CT images are imported into the treatment planning system. It also explains how the treatment planning system localizes CT markers and defines the isocenter in machine coordinates for treatment. Finally, it summarizes the verification process of aligning the patient using digital reconstructed radiographs and portal images to ensure accurate treatment delivery.
TBI is the radiotherapy technique to irradiate whole body before doing stem cell transplant. The main purpose of doing TBIB is to condition the immune system of body so that there will be maximum chance of transplant acceptance.
This document discusses techniques for simulation, planning, and treatment delivery for stereotactic body radiation therapy (SBRT) for liver metastases. It covers important steps including patient preparation, positioning, immobilization, motion management, imaging, and treatment execution. Motion management techniques discussed include abdominal compression, breath holding, gating, and tracking using internal or external surrogates. The importance of accurate simulation and reproducibility is emphasized for precise SBRT delivery.
This document discusses the importance of treatment verification in radiotherapy and outlines the process. It notes that even small errors can have negative consequences so treatment verification is essential to ensure the right dose is delivered to the right area. The key aspects of treatment verification are machine setup, monitor units, patient positioning and imaging by comparing images to references. Errors can be systematic from planning or random from daily variations; various methods are described to reduce errors and ensure treatments are accurately delivered.
Final ICRU 62 ( International commission on radiation units and measurements)DrAyush Garg
The document discusses recommendations from reports by the International Commission on Radiation Units and Measurements (ICRU) for defining volumes used in radiation therapy planning and reporting. ICRU Report 62 provides additional details on volumes such as the internal target volume (ITV) and planning organ at risk volume (PRV), and introduces metrics like the conformity index. It also further classifies organs at risk as serial, parallel or serial-parallel based on their radiosensitivity.
The vmat vs other recent radiotherapy techniquesM'dee Phechudi
VMAT is a new type of intensity-modulated radiation therapy (IMRT) treatment technique that uses the same hardware (i.e. a digital linear accelerator) as used for IMRT or conformal treatment, but delivers the radiotherapy treatment using a rotational or arc geometry rather than several static beams.
This technique uses continuous modulation (i.e. moving the collimator leaves) of the multileaf collimator (MLC) fields, continuous change of the fluence rate (the intensity of the X rays) and gantry rotation speed across a single or multiple 360 degree rotations
This document discusses image-guided radiation therapy (IGRT) and its evolution and applications. It begins by defining IGRT as external beam radiation therapy using imaging prior to each treatment fraction to verify patient positioning. IGRT allows for reduction of safety margins by compensating for set-up errors and organ motion. The document then reviews the history of IGRT from early portal imaging to modern cone-beam CT and other volumetric imaging techniques. It provides examples of IGRT protocols and clinical outcomes for sites such as prostate, lung, liver, and central nervous system tumors.
The document discusses the use of Tomotherapy for radiation treatment planning and delivery. It provides examples of how Tomotherapy allows for:
1) Highly conformal radiation plans that sculpt dose around complex tumor target shapes while minimizing dose to nearby organs.
2) Daily image guidance that enables adjustment of targets to account for changes in patient anatomy and tumor size during treatment.
3) Delivery of simultaneous integrated boosts to multiple tumor sites.
This document discusses 5 reasons why radiology needs artificial intelligence:
1. There is a global shortage of radiologists that is expected to worsen as imaging volumes increase faster than new radiologists enter the field. AI can help improve radiologist productivity.
2. AI can enhance radiologist productivity through functions like smart alerts, automatic image annotation, and faster access to patient information.
3. AI can help improve diagnostic accuracy by reducing human error, tracking lesions over time more quantitatively, and providing automated second opinions.
4. AI may help lower misdiagnosis rates by alerting radiologists to disease indicators and comparing new cases to existing ones to reduce fatigue-related errors.
5. AI could help improve
An overview of imrt, vmat optimization algorithmsRahim Gohar
This document discusses intensity-modulated radiation therapy (IMRT) and volumetric modulated arc therapy (VMAT) optimization algorithms, including direct machine parameter optimization (DVO), leaf motion calculation (LMC), progressive gradient optimization (PGO), and multi-resolution dose calculation (MRDC). It emphasizes that theoretical physicists developing these algorithms have revolutionized the field of radiation oncology by enabling more conformal radiation treatments.
This document provides an overview of image-guided radiation therapy (IGRT) for lung cancer. It discusses the role of IGRT in managing tumor motion through techniques like breath hold methods, free breathing with gating or tracking, and 4D imaging. Segmentation of the tumor and organs at risk on 4D CT scans is covered. Dose fractionation schedules and biological effective dose calculations for hypofractionated stereotactic body radiation therapy are reviewed. Toxicities, outcomes, and challenges of IGRT in lung cancer are also mentioned.
Basic information about Elekta and its familiar with xvi and Iviewgt protocols and there import and defining the Target area clip box registration along with HEXAPOD 6Dof couch & Apex Dmlc setup
This document discusses artificial intelligence and its applications in radiology. It begins with definitions of artificial intelligence and its subsets of machine learning and deep learning. It then discusses how machine learning and deep learning are being used in medical imaging for tasks like cancer diagnosis and detection of findings in images. The document outlines how large amounts of medical image and patient data are being used to train AI models to perform tasks like segmentation and anomaly detection. It provides examples of startups and projects applying AI to problems in radiology. It concludes by discussing views on AI in radiology, noting that AI can increase radiologist efficiency and consistency if integrated into the workflow, rather than replacing radiologists.
This document summarizes the expectations and key learnings from a linear accelerator acceptance, commissioning, and annual QA training that occurred from September to November 2008. The training covered:
1. Fundamental concepts of linear accelerators, beam production, safety features, and the acceptance testing process.
2. Techniques for collecting beam data needed for commissioning, including measurements and data definitions.
3. Procedures for linear accelerator QA and other treatment machine QA on an annual basis.
Key topics included the beamline components that produce photon and electron beams, characteristics of linear accelerator beams, the importance of acceptance testing and commissioning the machine properly, and techniques for annual QA tests.
Three dimensional conformal radiotherapy - 3D-CRT and IMRT - Intensity modula...Abhishek Soni
Conformal radiation therapy techniques like 3D CRT and IMRT aim to concentrate radiation dose in the tumor while sparing surrounding normal tissues. This is achieved through advances in imaging, treatment planning and delivery. 3D CRT uses geometric field shaping with multiple beams while IMRT further modulates beam intensity across each field. Both require contouring of target and organs at risk on imaging along with inverse or forward treatment planning to optimize dose distribution. Conformal techniques allow higher tumor doses with improved normal tissue sparing compared to conventional radiation therapy.
Artificial Intelligence in Radiation OncologyWookjin Choi
The document discusses artificial intelligence applications in radiation oncology. It begins with acknowledgements and then outlines topics including radiomics decision support tools, automatic delineation and variability analysis, and applications like lung cancer screening, tumor response prediction, and aggressive lung adenocarcinoma subtype prediction. Radiomics frameworks and deep learning models are presented. Results show potential for AI to provide quantitative imaging biomarkers and improve outcomes in areas like screening, treatment planning, and response assessment.
Radiomics Analysis of Pulmonary Nodules in Low Dose CT for Early Detection of...Wookjin Choi
Purpose: To predict the histopathologic subtypes with poor surgery prognosis in early stage lung adenocarcinomas using CT and PET radiomics.
Methods: We retrospectively enrolled 53 patients with stage I lung adenocarcinoma who underwent both diagnostic CT and 18F-fluorodeoxyglucose (FDG) PET/CT before complete surgical resection of the tumors. Tumor segmentation was manually contoured by a physician on both the diagnostic CT and the attenuation CT of PET/CT.A total of 170 radiomics features were extracted on both PET and CT images to design predictive models for two histopathologic endpoints: (1) tumors with solid or micropapillary predominant subtype (aggressiveness), and (2) tumors with micropapillary component more than 5% (MIP5). We used least absolute shrinkage and selection operator (LASSO) as a model building method coupled with a class separability feature selection (CSFS) method. For an unbiased model estimate, a 10-fold cross validation approach was used. The area under the curve (AUC) and prediction accuracy were employed to evaluate the performance of the model. P-values were computed using Wilcoxon rank-sum test.
Results: Of the 53 patients, 9 and 15 had tumors with aggressiveness and MIP5, respectively. For both endpoints, LASSO models with two PET radiomics features achieved the best performance. For aggressiveness, the LASSO model with PET Cluster Shade and PET 2D Variance resulted in 77.6±2.3% accuracy and 0.71±0.02 AUC (P = 0.011). For MIP5, the LASSO model with PET Eccentricity and PET Cluster Shade resulted in 69.6±3.1% accuracy and 0.68±0.04 AUC (P=0.014). The PET Cluster Shade was commonly selected in both models. Cluster shade is a texture feature that measures the skewness of the co-occurrence matrix. Higher PET cluster shade predicted that the tumor was more aggressive and more likely MIP5.
Conclusion: We showed that PET/CT radiomics features can predict tumor aggressiveness.
Funding Support, Disclosures, and Conflict of Interest: This work was supported in part by the National Cancer Institute Grants R01CA172638.
Radiomics and Deep Learning for Lung Cancer ScreeningWookjin Choi
The document summarizes research on using radiomics and deep learning approaches for lung cancer screening. It describes:
1) Using radiomic features like shape, texture, and intensity from lung nodules on CT scans and an SVM-LASSO model to classify nodules with 87.9% sensitivity and 78.2% specificity, outperforming the Lung-RADS system.
2) A deep learning model developed for a Kaggle competition that achieved 67.4% accuracy on nodule classification but only ranked 99th due to overfitting issues without enough data.
3) Future work could integrate quantification of nodule characteristics like spiculation with plasma biomarkers to improve diagnostic accuracy.
The Cancer Imaging Archive (TCIA) is a large online archive of medical images and associated clinical data from cancer patients. It contains a variety of imaging modalities like CT, MRI, and PET scans covering many cancer types. The archive aims to support precision medicine by linking imaging data to molecular and genomic data from sources like The Cancer Genome Atlas. It provides a growing collection of over 40,000 subjects and 70 datasets that are frequently used in research publications and challenge competitions. The TCIA helps relieve researchers of data sharing burdens and provides hosting, de-identification, and support services to submitters and users of the archive.
Prostate Cancer Diagnosis using Deep Learning with 3D Multiparametric MRI: Pr...Saifeng (Aaron) Liu
This document discusses a deep learning approach for predicting prostate cancer Gleason Grade Group from multiparametric MRI data. The authors developed a model called SummerNet that uses a multi-level approach, combining deep learning and conventional machine learning. SummerNet achieved a Cohen's kappa of 0.26 on the test data of the PROSTATEx-2 challenge, outperforming models that only used deep learning or conventional machine learning. The authors conclude that combining deep and conventional machine learning can help reduce overfitting, and suggest directions for future improvement including using multi-channel 3D convolutional layers.
American Association for Cancer Research Annual Meeting 2022
Analysis of images of routinely acquired tissue specimens promise to provide biomarkers that can be used to predict disease outcome and steer treatment, improve diagnostic reproducibility, and reveal new insights to further advance current human understanding of disease. The advent of AI and ubiquitous high-end computing are making it possible to carry out accurate whole slide image morphological and molecular tissue analyses at cellular and subcellular resolutions. AI methods are can enable exploration and discovery of novel diagnostic biomarkers grounded in prognostically predictive spatial and molecular patterns as well as quantitative assessments of predictive value and reproducibility of traditional morphological patterns employed in anatomic pathology. AI methods may be adapted to help steer treatment through integrative analysis of clinical information along with Pathology, Radiology and molecular data.
Longitudinal Plasma Samples: Paving the Way for Precision OncologyInsideScientific
Experts present a cell-free plasma biobank and describe the role of longitudinal plasma samples for cancer research, disease monitoring, and biomarker development.
Through liquid biopsies, it is now possible to repeatedly and non-invasively interrogate the molecular landscape of solid tumors via a blood draw over the whole treatment course. Until now, liquid biopsies can be used for screening, disease monitoring and prognosis. Circulating tumor DNA (ctDNA) and circulating tumor cells (CTCs) have been the most explored targets in this technology for commercial applications up to the present time.
In collaboration with a continuously expanding oncology network, Indivumed Services has established a unique high-quality cell-free plasma biobank that is exclusively focused on collecting longitudinal whole blood samples from cancer patients. This allows molecular insight by providing quick access to longitudinal plasma from cancer patients that have undergone treatment. ctDNA can then be isolated from longitudinal cell-free plasma to allow for monitoring of disease progression by providing diagnostic and prognostic information, potentially in real time.
Key Topics Include:
- Gain insights into Indivumed Services’ longitudinal plasma collection process
- Understand the advantages and benefits of utilizing longitudinal plasma samples for cancer research
- Explore applications of longitudinal plasma samples for biomarker research and development of companion diagnostics
Longitudinal Plasma Samples: Paving the Way for Precision OncologyInsideScientific
Experts present a cell-free plasma biobank and describe the role of longitudinal plasma samples for cancer research, disease monitoring, and biomarker development.
Through liquid biopsies, it is now possible to repeatedly and non-invasively interrogate the molecular landscape of solid tumors via a blood draw over the whole treatment course. Until now, liquid biopsies can be used for screening, disease monitoring and prognosis. Circulating tumor DNA (ctDNA) and circulating tumor cells (CTCs) have been the most explored targets in this technology for commercial applications up to the present time.
In collaboration with a continuously expanding oncology network, Indivumed Services has established a unique high-quality cell-free plasma biobank that is exclusively focused on collecting longitudinal whole blood samples from cancer patients. This allows molecular insight by providing quick access to longitudinal plasma from cancer patients that have undergone treatment. ctDNA can then be isolated from longitudinal cell-free plasma to allow for monitoring of disease progression by providing diagnostic and prognostic information, potentially in real time.
Key Topics Include:
- Gain insights into Indivumed Services’ longitudinal plasma collection process
- Understand the advantages and benefits of utilizing longitudinal plasma samples for cancer research
- Explore applications of longitudinal plasma samples for biomarker research and development of companion diagnostics
Interpretable Spiculation Quantification for Lung Cancer ScreeningWookjin Choi
Spiculations are spikes on the surface of pulmonary nodule and are important predictors of malignancy in lung cancer. In this work, we introduced an interpretable, parameter-free technique for quantifying this critical feature using the area distortion metric from the spherical conformal (angle-preserving) parameterization. The conformal factor in the spherical mapping formulation provides a direct measure of spiculation which can be used to detect spikes and compute spike heights for geometrically-complex spiculations. The use of the area distortion metric from conformal mapping has never been exploited before in this context. Based on the area distortion metric and the spiculation height, we introduced a novel spiculation score. A combination of our spiculation measures was found to be highly correlated (Spearman's rank correlation coefficient ρ = 0.48) with the radiologist's spiculation score. These measures were also used in the radiomics framework to achieve state-of-the-art malignancy prediction accuracy of 88.9% on a publicly available dataset.
Dekker trog - learning outcome prediction models from cancer data - 2017Andre Dekker
The document discusses learning outcome prediction models from cancer data. It begins by outlining the speaker's disclosures and research collaborations. It then lists the talks being given at the conference related to learning outcome prediction models, big data in radiation oncology, and knowledge engineering in oncology. The objectives of the talk are then stated as understanding sources of cancer data and challenges of sharing data, the methodology to develop prediction models from data, and how to appraise papers describing models.
Utilization of NGS to Identify Clinically-Relevant Mutations in cfDNA: Meet t...QIAGEN
Pancreatic cancer is a uniquely lethal malignancy characterized by frequent mutations in KRAS, CDKN2A, SMAD4, TP53 and many others. We have shown that KRAS mutation can be detected in cell-free, circulating tumor DNA (ctDNA) isolated from the plasma in a subset of patients and is associated with poor prognosis. The ability to simultaneously detect multiple pancreatic cancer-specific mutations in ctDNA would open a new avenue for detection of clinically-relevant mutations. In this study, we performed ultra-deep sequencing of ctDNA from advanced pancreatic cancer patients prior to treatment with Gemcitabine and Erlotinib following target enrichment. Somatic, non-synonymous variants were identified in 29 different genes at allele frequencies typically less than 0.5%. Updated results of ultra-deep NGS analysis will be presented.
Pathomics Based Biomarkers, Tools, and Methodsimgcommcall
This document discusses pathomics-based biomarkers, tools, and methods for multi-scale integrative analysis in biomedical informatics. It summarizes several projects involving extracting quantitative features from pathology and radiology images using image segmentation and analysis techniques. These features are then linked to molecular data and clinical outcomes using statistical and machine learning methods to develop biomarkers. The tools and methods described aim to standardize and optimize feature extraction while accounting for uncertainties.
Dekker trog - big data for radiation oncology - 2017Andre Dekker
- Big data in radiation oncology comes from clinical research, registries, and routine clinical data, but the latter has the most patients and features while also having the most missing data.
- Models are developed using a hypothesis-driven approach by learning from a training cohort and estimating performance in a validation cohort. Challenges include gaining trust in models, dealing with continuous changes, and addressing barriers to implementing shared decision making.
- Overall, big data and models can improve cancer care by better tailoring treatments to individual patients, but also require overcoming challenges through rapid learning and collaboration across institutions.
IMAGING BIOMARKER PANELS AND MULTI-OMICS AI MODELS FOR OUTCOMES PREDICTIONiQHub
Quibim is a company that develops artificial intelligence models and multi-omics biomarker panels to predict clinical outcomes using medical imaging data. It has over 150 journal publications, treats patients at 80+ sites, and has FDA/CE/UKCA clearances. Quibim aims to improve disease understanding, early detection of at-risk patients, and identification of likely treatment responders through its personalized medicine approaches. Key challenges include validating AI models across sites and generating sufficient evidence to impact clinical guidelines. Quibim addresses these by harmonizing heterogeneous imaging data, automating organ/lesion detection, and establishing partnerships to build large predictive imaging biomarker panels.
Oncology is a branch of medicine that deals with the prevention, diagnosis and treatment of cancer. Research contains mechanism study (including molecule mechanism), pathological research, surgical method, medical treatment therapy (chemotherapy, radiotherapy), management for an individual patient, new drug development, etc. Obviously, oncology is often managed through discussion on multi-disciplinary.
https://www.creative-bioarray.com/oncology.htm
Deep Learning-based Histological SegmentationDifferentiates Cavitation Patte...Wookjin Choi
Unsupervised segmentation (unlabeled regions of interest, ROIs) and autoencoder (AE)-based classification were used to classify differences in cavitation patterns in knees and digits using the stained images (n=20-30 images/group).
Each image was divided into 256 x 256 pixel patches, and a convolutional neural network (CNN)-based unsupervised segmentation was used to identify ROIs. These patches were subsequently fed into a CNN-based AE whose latent space layer was connected to a classifier for input patch classification.
The AE was trained using the ROIs identified by the unsupervised segmentation, and the image classes were used to train the classifier. Whole image classifications were determined by maximum voting of the patch results and evaluated by accuracy.
Artificial Intelligence To Reduce Radiation-induced Cardiotoxicity In Lung Ca...Wookjin Choi
Traditionally, radiation-induced cardiotoxicity has been studied using cardiac radiation doses rather than functional imaging. We developed artificial intelligence (AI) models based on novel cardiac delta radiomics using pre- and post-treatment FDG-PET/CT scans to predict overall survival in lung cancer patients undergoing radiotherapy. We identified four clinically relevant delta radiomics features with the AI prediction models. The best model achieved an AUC of 0.91 on the training set and 0.87 on the test set. We are a pioneering group in AI for functional cardiac imaging. If validated, this approach will enable to use standard PET/CT scans as functional cardiac imaging with good predictive AUC for OS, as well as provide automated methods to provide functional cardiac information for clinical outcome prediction AI in lung cancer patients.
Novel Functional Radiomics for Prediction of Cardiac PET Avidity in Lung Canc...Wookjin Choi
Purpose/Objective(s)
Traditional methods of evaluating cardiotoxicity focus solely on radiation doses to the heart and do not incorporate functional imaging information. Functional imaging has great potential to improve the ability to provide early prediction for cardiotoxicity for lung cancer patients undergoing radiotherapy. FDG-based PET/CT imaging is routinely obtained as part of standard staging work up for lung cancer patients. Although FDG PET/CT scans are typically used to evaluate the tumor, imaging guidelines note that FDG PET/CT scans are an FDA-approved method to image for cardiac inflammation, and studies have noted that the PET cardiac signal can be predictive of clinical outcomes. The purpose of this work was to develop a radiomics model to predict clinical cardiac assessment of standard of care FDG PET/CT scans.
Materials/Methods
The study included 100 consecutive lung cancer patients treated with radiotherapy who underwent standard pre-treatment FDG-PET/CT staging scans. A clinician reviewed the PET/CT scans per clinical cardiac assessment guidelines and classified the cardiac uptake as: 0 = uniform diffuse, 1 = absent, 2 = heterogeneous, with event rates of 20%, 44%, and 35%, respectively. The heart was delineated and 200 novel functional radiomics features were selected to classify cardiac FDG uptake patterns. We divided the data into an 80% training set and a 20% test set to train and evaluate the classification models. Feature reduction was carried out using the Wilcoxon test (with Bonferroni adjusted p<0.05), hierarchical clustering, and Recursive Feature Elimination. Two automatic machine learning (AutoML) frameworks were used to determine classification models: a Random Forest Classifier (Tree-based Pipeline Optimization Tool, TPOT) and Linear Discriminant Analysis (AutoSklearn). 10-fold cross validation was carried out for training and the accuracy of the ability of the models to predict for clinical cardiac assessment is reported.
Results
Fifty-one independent radiomics features were reduced to 3 clinically pertinent features (PET 2D Skewness, PET Grey Level Co-occurrence Matrix Correlation, and PET Median) using feature reduction techniques. The model selected by TPOT showed 89.8% predictive accuracy in the cross validation of the training set and 85% predictive accuracy on the test set. The model selected by AutoSklearn showed 89.7% predictive accuracy in the cross validation of the training set and 80% predictive accuracy on the test set.
Conclusion
The novelty of this work is that it is the first study to develop and evaluate functional cardiac radiomic features from standard of care FDG PET/CT scans with the data showing good predictive accuracy with clinical imaging evaluation. If validated, the current work provides automated methods to provide functional cardiac information using standard of care imaging that can be used as an imaging biomarker for early clinical toxicity prediction for lung cancer patients.
Novel Deep Learning Segmentation Models for Accurate GTV and OAR Segmentation...Wookjin Choi
Purpose/Objective(s)
MR-guided adaptive radiotherapy (MRgART) improves target coverage and organ-at-risk (OAR) sparing in pancreatic cancer radiation therapy (RT). Inter-fractional changes in patients undergoing RT require time intensive re-delineation of gross tumor volume (GTV) and OARs prior to adaptive optimization. Accurate automatic segmentation has the potential to significantly improve efficiency of the adaptive workflow. We hypothesized that state-of-the-art deep learning (DL) segmentation models could adequately segment GTV and OARs in both planning and daily fractional MR scans.
Materials/Methods
The study included 21 patients with pancreatic cancer treated with MRgART (10 Gy x 5 fractions). The planning MR as well as all daily MR images and registrations were collected (6 image sets per patient and a total of 126 image sets). The planning MR and fraction 1-4 image sets were used as the training set (N = 105), while the test set (N = 21) comprised images for fraction 5, to simulate the last step of incremental learning from planning to final fraction. Evaluated contours included the GTV, Small Bowel, Large Bowel, Duodenum, Left and Right Kidney, Liver, Spinal Cord, and Stomach. To mimic clinical conditions, contour accuracy was evaluated within the ring structure surrounding the PTV, inside of which daily adaptive re-contouring is applied (2 cm expansion in the cradio-caudal direction, 3 cm expansion otherwise). We evaluated three DL model architectures: SegResNet, SegResNet 2D, and SwinUNETR to autosegment GTV and OARs. The segmentation models were trained on the training set using 5-fold cross-validation (CV) and quantitatively analyzed by comparing against clinically used contours with DICE scores. Qualitative analysis was performed by a radiation oncologist using a scoring scale: 1 = perfect, 2 = minor discrepancy, 3 = moderate discrepancy, and 4 = rejected.
Results
Overall, the DL segmentations were in acceptable agreement with clinical contours. The best performing model was the SwinUNETR model with overall training DICE = 0.88±0.06, test DICE = 0.78±0.11, and qualitative score of 1.6±0.8. The agreement between the DL model and clinical segmentation for the GTV was 0.79±0.08, with a qualitative score of 2.2±0.9
Conclusion
We report here the most comprehensive work on DL segmentation for pancreatic cancer MRgART, including quantitative and clinically-pertinent qualitative evaluations of 126 image sets and 3 DL architectures. Our data show good quantitative agreement between DL and clinical contours, and acceptable clinician evaluations for the majority of GTVs and OARs. The current work has great potential to significantly reduce a major bottleneck in the MRgART workflow for pancreatic cancer patients.
Novel Functional Delta-Radiomics for Predicting Overall Survival in Lung Canc...Wookjin Choi
AAPM2023_SU-300-IePD-F6-4
Purpose: Traditional methods of evaluating cardiotoxicity rely on cardiac radiation doses and do not incorporate functional imaging. Cardiac functional imaging can improve the ability to provide early prediction for clinical outcomes for lung cancer patients undergoing radiotherapy. FDG-based PET/CT imaging is routinely obtained for staging and disease assessment after treatment. Although FDG PET/CT scans are typically used to evaluate the tumor, studies have shown that the PET cardiac signal is predictive of clinical outcomes. Our study aimed to develop novel functional cardiac delta radiomics using pre and post-treatment FDG PET/CT scans to predict for overall survival (OS).
Methods: We conducted a study of 109 lung cancer patients who underwent standard FDG-PET/CT scans pre- and post-radiotherapy. Data from ACRIN 6668 (N=70) and an investigator-initiated lung cancer trial (N=39) for functional avoidance radiotherapy were used. The heart was delineated, and 200 cardiac CT and PET functional radiomics features were selected. Delta radiomics was calculated as the change between pre- and post-PET/CT. The data were divided into 80%/20% training/test set, and feature reduction was performed using Wilcoxon test, hierarchical clustering, and recursive feature elimination. A Gradient Boosting Classifier machine learning model evaluated the ability of the delta PET/CT cardiac radiomics to predict for OS using 10-fold cross-validation for training and area-under-the-curve (AUC) for model assessment.
Results: Median survival was 431 days (range 144 to 1640 days). 4 clinically relevant delta features were identified: pre-CT_Maximum, post-CT_Minimum, delta-CT_GLRM_Run_Variance, delta-PET_GLRM_Run_Entropy. The model showed an AUC of 0.91 on the training set and an AUC of 0.87 on the test set.
Conclusion: This is the first study to evaluate functional cardiac delta radiomic features from standard PET/CT scans with data showing good predictive AUC for OS. If validated, this work provides automated methods to provide functional cardiac information for clinical outcome prediction in lung cancer patients.
CIRDataset: A large-scale Dataset for Clinically-Interpretable lung nodule Ra...Wookjin Choi
The CIRDataset provides a large-scale dataset of 956 annotated lung nodules with segmentations and classifications of spiculations and lobulations, which are important radiomic features for assessing malignancy. It aims to address the lack of publicly available datasets capturing these subtle radiological features typically assessed by radiologists but often smoothed over by deep learning segmentation models. The dataset is accompanied by code, models, and a pipeline to enable the development of AI systems for joint nodule segmentation, classification of spiculations/lobulations, and malignancy prediction using an end-to-end deep learning approach.
Automatic motion tracking system for analysis of insect behaviorWookjin Choi
Undergraduate research.
We present a multi-object tracking system to track small insects such as ants and bees. Motion-based object tracking recognizes the movements of objects in videos using information extracted from the given video frames. We applied several computer vision techniques, such as blob detection and appearance matching, to track ants. Moreover, we discussed different object detection methodologies and investigated the various challenges of object detection, such as illumination variations and blob merge/split. The proposed system effectively tracked multiple objects in various environments.
Assessing the Dosimetric Links between Organ-At-Risk Delineation Variability ...Wookjin Choi
Purpose: To determine the relative dosimetric impact of delineation variability (DV) when inter-observer and inter-technique planning variability (PV), and setup variability (SV) with are considered.
Methods: 409 plans for a single head-and-neck patient from the 2017 Radiation Knowledge plan competition were used. Plans were created with Eclipse (N=227), Pinnacle (N=49), RayStation (N=25), Monaco (N=75), and TomoTherapy (N=33) with delivery techniques conventional linac IMRT (N=142), volumetric modulated arc therapy (VMAT, N=234), and helical TomoTherapy (N=33). All plans were optimized using a consistent set of target volumes and a single OAR structure set. Four additional OAR structure sets were contoured by radiation oncologists (N=2) and medical physics residents (N=2) who had completed head-and-neck contouring training. Probabilistic DVHs, dose-volume coverage maps (DVCM), which shows the probability of achieving a dose metric, were computed for each OAR on the following scenarios: SV alone (N=1000), SV+PV (N=1000*409), SV+DV (N=1000*5), SV+PV+DV (total variability [TV], N=1000*409*5). Analysis focused on the probability of exceeding the maximum dose constraint exceeded 5% for each OAR.
Results: The primary source of variability was PV, which was expected due to inter-observer planning abilities and preferences during the optimization planning process, even when all participants utilized the same constraints. The parotid had the most significant interquartile range (IQR) on the PV scenario. Conversely, adding SV, DV, and TV each reduced the IQR, showing a washing out effect on the DVCM.
Conclusion: Assessment of OAR sensitivity to DV will be highly sensitive to the specific planning technique and planner, likely requiring plan-specific assessment of in-tolerance delineation variations. Incorporation SV and DV variabilities in plan assessments washes out their relative impacts on maximum dose.
Simulation of Realistic Organ-At-Risk Delineation Variability in Head and Nec...Wookjin Choi
(Sunday, 7/14/2019) 4:00 PM - 5:00 PM
Room: 225BCD
Purpose: To simulate realistic manual delineation (MD) organ-at-risk (OAR) delineation variability (DV) the purpose of quantifying DV’s dosimetric impact.
Methods: Fourteen independent MD head-and-neck OAR structure sets (SS) were obtained from the ESTRO Falcon group. Seven OARs were available (BrainStem, Esophagus, OralCavity, Parotid_L, Parotid_R, SpinalCord, and Thyroid). A consensus MD SS was generated by the simultaneous truth and performance level estimation (STAPLE) method. MD DV was evaluated with respect to the STAPLE SS using the Dice coefficient and Hausdorff distance (HD) geometric similarity metrics. DVs were simulated using auto-delineation (AD)
methods: an average surface of standard deviation (ASSD) method, GrowCut segmentation, and a random walker (RW) segmentation. Each OAR AD was repeated five times with a different seed or variability level. Dice and HD were computed for each OAR AD with respect to the STAPLE SS. Dosimetric analysis was achieved by intercomparing dose-volume histograms (DVH) from a plan developed with a reference MD SS with DVHs for each MD and AD. DVH confidence bands are reported for MD and each AD method.
Results: The MD Dice was 0.7±0.2 (μ±σ). AD Dice values (ASSD, GrowCut, and RW) were 0.5±0.2, 0.7±0.2, and 0.8±0.1, respectively. HDs were 35.4±45.2, 27.3±19.1, 29.3±19.9, and 14.6±10.3. The simulated DV increased with increasing the seed standard deviations or variability level. The dosimetric effect was largest for MD DVs (larger OAR DVH confidence intervals and larger HD), even though the MD Dice was greater than the ASSD and GrowCut Dice values. GrowCut DV resulted in less dosimetric variation than RW, unlike the geometric indices.
Conclusion: We developed a framework to simulate DVs and demonstrated its feasibility. ADs were able to simulate different magnitudes of DVs, but did not replicate the dosimetric consequences of human delineation variability. The correlation between geometric similarity metrics and dosimetric consequences of DV is poor.
Quantitative image analysis for cancer diagnosis and radiation therapyWookjin Choi
1.Lung Cancer Screening
1.1.Deep learning (feasible but not interpretable)
1.2.Radiomics (concise model)
1.3.Spiculation quantification (interpretable feature)
2.PET/CT Tumor Response
2.1.Aggressive Lung ADC subtype prediction (helpful for surgeons)
2.2.Pathologic response prediction (accurate but not concise)
2.3.Local tumor morphological changes (accurate and interpretable)
Interpretable Spiculation Quantification for Lung Cancer ScreeningWookjin Choi
Spiculations are spikes on the surface of pulmonary nodule and are important predictors of malignancy in lung cancer. In this work, we introduced an interpretable, parameter-free technique for quantifying this critical feature using the area distortion metric from the spherical conformal (angle-preserving) parameterization. The conformal factor in the spherical mapping formulation provides a direct measure of spiculation which can be used to detect spikes and compute spike heights for geometrically-complex spiculations. The use of the area distortion metric from conformal mapping has never been exploited before in this context. Based on the area distortion metric and the spiculation height, we introduced a novel spiculation score. A combination of our spiculation measures was found to be highly correlated (Spearman's rank correlation coefficient ρ = 0.48) with the radiologist's spiculation score. These measures were also used in the radiomics framework to achieve state-of-the-art malignancy prediction accuracy of 88.9% on a publicly available dataset.
Quantitative Image Analysis for Cancer Diagnosis and Radiation TherapyWookjin Choi
1.Lung Cancer Screening
1.1.Deep learning (feasible but not interpretable)
1.2.Radiomics (concise model)
1.3.Spiculation quantification (interpretable feature)
2.PET/CT Tumor Response
2.1.Aggressive Lung ADC subtype prediction (helpful for surgeons)
2.2.Pathologic response prediction (accurate but not concise)
2.3.Local tumor morphological changes (accurate and interpretable)
Robust Normal Lung CT Texture Features for the Prediction of Radiation-Induce...Wookjin Choi
Abstract
Purpose: Normal lung CT texture features have been used for the prediction of radiation-induced lung disease (RILD) - pneumonitis and fibrosis. For these features to be clinically useful, they should be robust (relatively invariant or unbiased) to tumor size variations and not correlated (non-redundant) with the normal lung volume of interest, i.e., volume of the peri-tumoral region.
Methods: CT images of 14 lung cancer patients were studied. Different sizes of gross tumor volumes (GTVs) were simulated with spheres (diameters 10 to 60 mm) and placed in the lung contralateral to the tumor. 27 texture features [nine from intensity histogram, eight from the gray-level co-occurrence matrix (GLCM) and ten from the gray-level run-length matrix (GLRM)] were extracted from the peri-tumoral region (uniform 30 mm expansion around the GTV in the lung). The Bland-Altman analysis was applied to measure the normalized range of agreement (nRoA) for each feature when GTV size varied. A feature was considered as robust when its nRoA was less than the threshold (100%), which was chosen at the nRoA of the volume of the peri-tumoral region with modification based on the cumulative graph of features vs. nRoA. A feature was regarded as not correlated with the volume of the peri-tumoral region when their correlation was lower than 0.70.
Results: 16 of the 27 texture features were identified as robust. All intensity histogram features were robust except sum and kurtosis. All GLCM features were robust except cluster shade, cluster prominence, and Haralick's Correlation. Five GLRM features (two run emphasis and three high gray-level emphasis) were robust while the other five (two nonuniformity and three low gray-level emphasis) were unrobost. None of the robust features was correlated with the volume of the peri-tumoral region. No feature showed statistically significant differences (P<0.05) on GTV location (upper vs. lower lobe).
Conclusion: We identified 16 robust normal lung CT texture features that can be further examined for the prediction of RILD. Particularly, GLRM high gray-level emphasis features were robust and characterized the radiologic manifestations of pulmonary abnormalities.
Robust Normal Lung CT Texture Features for the Prediction of Radiation-Induce...Wookjin Choi
Abstract
Purpose/Objective(s)
Normal lung CT texture features have been used for the prediction of radiation-induced lung disease (RILD) - pneumonitis and fibrosis. For these features to be clinically useful, they need to be relatively invariant (robust) to tumor size and not correlated with normal lung volume.
Materials/Methods
The free-breathing CTs of 14 lung SBRT patients were studied. Different sizes of GTVs were simulated with spheres (diameters 10 to 60 mm) placed in the lung contralateral to the tumor. Twenty-seven texture features (9 from intensity histogram, 8 from the gray-level co-occurrence matrix [GLCM], and 10 from the gray-level run-length matrix [GLRM]) were extracted from [lung – GTV]. The Bland-Altman method was applied to measure the normalized range of agreement (nRoA) of each texture feature when GTV size varied. A feature was considered as robust when its nRoA was less than that of [lung – GTV] volume (8.8%) and regarded as not correlated when their absolute correlation coefficient was lower than 0.70.
Results
Eighteen texture features were identified as robust. All intensity histogram features were robust except sum and kurtosis. All GLCM features were robust except energy and Haralick's Correlation. Five GLRM features (two run emphasis and three high gray-level emphasis) were robust while the other five (two nonuniformity and three low gray-level emphasis) were nonrobust. Particularly, all three low gray-level emphasis features had extremely large nRoAs (∼30%), indicating huge variations when GTV size changed. None of the robust features was correlated with the normal lung [lung – GTV] volume, suggesting that they can provide additional information. Three nonrobust features (sum and two nonuniformity features) were highly correlated with the normal lung volume. None feature showed statistically significant differences (P < 0.05) with respect to GTV location (upper vs. lower lobe).
Conclusion
We identified 18 robust lung CT texture features which were invariant to varying tumor volumes. Particularly the three GLRM high gray-level emphasis features can characterize the radiologic manifestations of pulmonary abnormalities. Hence these features can be further examined for the prediction of the RILD.
Aggressive Lung Adenocarcinoma Subtype Prediction Using FDG-PET/CT RadiomicsWookjin Choi
Purpose: To predict the histopathologic subtypes with poor surgery prognosis in early stage lung adenocarcinomas using CT and PET radiomics.
Methods: We retrospectively enrolled 53 patients with stage I lung adenocarcinoma who underwent both diagnostic CT and 18F-fluorodeoxyglucose (FDG) PET/CT before complete surgical resection of the tumors. Tumor segmentation was manually contoured by a physician on both the diagnostic CT and the attenuation CT of PET/CT.A total of 170 radiomics features were extracted on both PET and CT images to design predictive models for two histopathologic endpoints: (1) tumors with solid or micropapillary predominant subtype (aggressiveness), and (2) tumors with micropapillary component more than 5% (MIP5). We used least absolute shrinkage and selection operator (LASSO) as a model building method coupled with a class separability feature selection (CSFS) method. For an unbiased model estimate, a 10-fold cross validation approach was used. The area under the curve (AUC) and prediction accuracy were employed to evaluate the performance of the model. P-values were computed using Wilcoxon rank-sum test.
Results: Of the 53 patients, 9 and 15 had tumors with aggressiveness and MIP5, respectively. For both endpoints, LASSO models with two PET radiomics features achieved the best performance. For aggressiveness, the LASSO model with PET Cluster Shade and PET 2D Variance resulted in 77.6±2.3% accuracy and 0.71±0.02 AUC (P = 0.011). For MIP5, the LASSO model with PET Eccentricity and PET Cluster Shade resulted in 69.6±3.1% accuracy and 0.68±0.04 AUC (P=0.014). The PET Cluster Shade was commonly selected in both models. Cluster shade is a texture feature that measures the skewness of the co-occurrence matrix. Higher PET cluster shade predicted that the tumor was more aggressive and more likely MIP5.
Conclusion: We showed that PET/CT radiomics features can predict tumor aggressiveness.
Funding Support, Disclosures, and Conflict of Interest: This work was supported in part by the National Cancer Institute Grants R01CA172638.
Individually Optimized Contrast-Enhanced 4D-CT for Radiotherapy Simulation in...Wookjin Choi
Purpose/Objectives: To develop an individually optimized contrast-enhanced (CE) 4D-CT for radiotherapy simulation in pancreatic adenocarcinoma (PDA).
Materials/Methods: Ten PDA patients were enrolled and underwent three CT scans: a 4D-CT immediately following a CE 3D-CT, and an individually optimized CE 4D-CT using a test injection to estimate the peak contrast enhancement time and to optimize the delay time. Three physicians contoured the tumor and pancreatic tissues. We compared image quality scores, tumor volume, motion, image noise, tumor-to-pancreas contrast, and contrast-to- noise ratio (CNR) in the three CTs. We also evaluated inter-observer variations in contouring the tumor using simultaneous truth and performance level estimation (STAPLE).
Results: The average image quality scores for CE 3D-CT and CE 4D-CT were comparable (4.0 and 3.8, p=0.47), and both were significantly better than that for 4D-CT (2.6, p<0.001). The tumor-to- pancreas contrast in CE 3D-CT and CE 4D-CT were comparable (15.5 and 16.7 HU, p=0.71), and the later was significantly higher than that in 4D-CT (9.2 HU, p=0.03). Image noise in CE 3D-CT (12.5 HU) was significantly lower than that in CE 4D-CT (22.1 HU, p<0.001) and 4D-CT (19.4 HU, p=0.005). The CNR in CE 3D-CT and CE 4D-CT were comparable (1.4 and 0.8, p=0.23), and the former was significantly better than that in 4D-CT (0.6, p=0.04). The average tumor volume was smaller in CE 3D-CT (29.8 cm 3 ) and CE 4D-CT (22.8 cm 3 ) than in 4D-CT (42.0 cm 3 ), though the differences were not statistically significant. The tumor motion was comparable in 4D-CT and CE 4D-CT (7.2 and 6.2 mm, p=0.23). The inter-observer variations were comparable in CE 3D-CT and CE 4D-CT (Jaccard index 66.0% and 61.9%), and the former was significantly smaller than that of 4D-CT (55.6%, p=0.047).
Conclusions: The CE 4D-CT demonstrated largely comparable characteristics to the CE 3D-CT. It has high potential for simultaneously delineating the tumor and quantifying the tumor motion with a single scan.
Individually Optimized Contrast-Enhanced 4D-CT for Radiotherapy Simulation in...Wookjin Choi
To develop an individually optimized contrast-enhanced (CE) 4D-CT for radiotherapy simulation in pancreatic ductal adenocarcinomas (PDA).
http://scitation.aip.org/content/aapm/journal/medphys/43/6/10.1118/1.4958261
Identification of Robust Normal Lung CT Texture FeaturesWookjin Choi
Normal lung CT texture features have been used for the prediction of radiation-induced lung disease (radiation pneumonitis and radiation fibrosis). For these features to be clinically useful, they need to be relatively invariant (robust) to tumor size and not correlated with normal lung volume.
http://scitation.aip.org/content/aapm/journal/medphys/43/6/10.1118/1.4955803
Robust breathing signal extraction from cone beam CT projections based on ada...Wookjin Choi
This document summarizes a research paper that proposes a novel method for extracting breathing signals from cone beam CT projections without using external markers. The method uses an adaptive filtering technique to enhance weak oscillating structures in the Amsterdam Shroud image generated from the projections. A two-step optimization approach is then used to reveal the large-scale regularity of the breathing signals. Evaluation on 5 patient data sets found the new algorithm outperformed existing methods by extracting less noisy signals with errors of only -0.07±1.58 breaths per minute compared to reference signals. While results are promising, the study had a small data set and image quality remains limited.
Local Advanced Lung Cancer: Artificial Intelligence, Synergetics, Complex Sys...Oleg Kshivets
Overall life span (LS) was 1671.7±1721.6 days and cumulative 5YS reached 62.4%, 10 years – 50.4%, 20 years – 44.6%. 94 LCP lived more than 5 years without cancer (LS=2958.6±1723.6 days), 22 – more than 10 years (LS=5571±1841.8 days). 67 LCP died because of LC (LS=471.9±344 days). AT significantly improved 5YS (68% vs. 53.7%) (P=0.028 by log-rank test). Cox modeling displayed that 5YS of LCP significantly depended on: N0-N12, T3-4, blood cell circuit, cell ratio factors (ratio between cancer cells-CC and blood cells subpopulations), LC cell dynamics, recalcification time, heparin tolerance, prothrombin index, protein, AT, procedure type (P=0.000-0.031). Neural networks, genetic algorithm selection and bootstrap simulation revealed relationships between 5YS and N0-12 (rank=1), thrombocytes/CC (rank=2), segmented neutrophils/CC (3), eosinophils/CC (4), erythrocytes/CC (5), healthy cells/CC (6), lymphocytes/CC (7), stick neutrophils/CC (8), leucocytes/CC (9), monocytes/CC (10). Correct prediction of 5YS was 100% by neural networks computing (error=0.000; area under ROC curve=1.0).
Does Over-Masturbation Contribute to Chronic Prostatitis.pptxwalterHu5
In some case, your chronic prostatitis may be related to over-masturbation. Generally, natural medicine Diuretic and Anti-inflammatory Pill can help mee get a cure.
Histololgy of Female Reproductive System.pptxAyeshaZaid1
Dive into an in-depth exploration of the histological structure of female reproductive system with this comprehensive lecture. Presented by Dr. Ayesha Irfan, Assistant Professor of Anatomy, this presentation covers the Gross anatomy and functional histology of the female reproductive organs. Ideal for students, educators, and anyone interested in medical science, this lecture provides clear explanations, detailed diagrams, and valuable insights into female reproductive system. Enhance your knowledge and understanding of this essential aspect of human biology.
ABDOMINAL TRAUMA in pediatrics part one.drhasanrajab
Abdominal trauma in pediatrics refers to injuries or damage to the abdominal organs in children. It can occur due to various causes such as falls, motor vehicle accidents, sports-related injuries, and physical abuse. Children are more vulnerable to abdominal trauma due to their unique anatomical and physiological characteristics. Signs and symptoms include abdominal pain, tenderness, distension, vomiting, and signs of shock. Diagnosis involves physical examination, imaging studies, and laboratory tests. Management depends on the severity and may involve conservative treatment or surgical intervention. Prevention is crucial in reducing the incidence of abdominal trauma in children.
Rasamanikya is a excellent preparation in the field of Rasashastra, it is used in various Kushtha Roga, Shwasa, Vicharchika, Bhagandara, Vatarakta, and Phiranga Roga. In this article Preparation& Comparative analytical profile for both Formulationon i.e Rasamanikya prepared by Kushmanda swarasa & Churnodhaka Shodita Haratala. The study aims to provide insights into the comparative efficacy and analytical aspects of these formulations for enhanced therapeutic outcomes.
Osteoporosis - Definition , Evaluation and Management .pdfJim Jacob Roy
Osteoporosis is an increasing cause of morbidity among the elderly.
In this document , a brief outline of osteoporosis is given , including the risk factors of osteoporosis fractures , the indications for testing bone mineral density and the management of osteoporosis
These lecture slides, by Dr Sidra Arshad, offer a quick overview of the physiological basis of a normal electrocardiogram.
Learning objectives:
1. Define an electrocardiogram (ECG) and electrocardiography
2. Describe how dipoles generated by the heart produce the waveforms of the ECG
3. Describe the components of a normal electrocardiogram of a typical bipolar lead (limb II)
4. Differentiate between intervals and segments
5. Enlist some common indications for obtaining an ECG
6. Describe the flow of current around the heart during the cardiac cycle
7. Discuss the placement and polarity of the leads of electrocardiograph
8. Describe the normal electrocardiograms recorded from the limb leads and explain the physiological basis of the different records that are obtained
9. Define mean electrical vector (axis) of the heart and give the normal range
10. Define the mean QRS vector
11. Describe the axes of leads (hexagonal reference system)
12. Comprehend the vectorial analysis of the normal ECG
13. Determine the mean electrical axis of the ventricular QRS and appreciate the mean axis deviation
14. Explain the concepts of current of injury, J point, and their significance
Study Resources:
1. Chapter 11, Guyton and Hall Textbook of Medical Physiology, 14th edition
2. Chapter 9, Human Physiology - From Cells to Systems, Lauralee Sherwood, 9th edition
3. Chapter 29, Ganong’s Review of Medical Physiology, 26th edition
4. Electrocardiogram, StatPearls - https://www.ncbi.nlm.nih.gov/books/NBK549803/
5. ECG in Medical Practice by ABM Abdullah, 4th edition
6. Chapter 3, Cardiology Explained, https://www.ncbi.nlm.nih.gov/books/NBK2214/
7. ECG Basics, http://www.nataliescasebook.com/tag/e-c-g-basics
1. Artificial Intelligence
in Radiation Oncology
Wookjin Choi, PhD
Assistant Professor of Radiation Oncology
Sidney Kimmel Medical College at Thomas Jefferson University
Wookjin.Choi@Jefferson.edu
Mar 11, 2022 @ Mayo Clinic
2. Acknowledgements
Memorial Sloan Kettering Cancer Center
• Wei Lu PhD
• Sadegh Riyahi, PhD
• Jung Hun Oh, PhD
• Saad Nadeem, PhD
• Eric Aliotta, PhD
• Joseph O. Deasy, PhD
• Andreas Rimner, MD
• Prasad Adusumilli, MD
Stony Brook University
• Allen Tannenbaum, PhD
University of Virginia School of Medicine
• Jeffrey Siebers, PhD
• Victor Gabriel Leandro Alves, PhD
University of Maryland School of Medicine
• Howard Zhang, PhD
• Wengen Chen, MD, PhD
• Charles White, MD
Thomas Jefferson University
• Yevgeniy Vinogradskiy, PhD
• Hamidreza Nourzadeh, PhD
• Adam P. Dicker, MD
2
NIH/NCI Grant R01 CA222216, R01 CA172638 and
NIH/NCI Cancer Center Support Grant P30 CA008748 and 5P30 CA056036
The ESTRO Falcon project team, Scott Kaylor of EduCase, Benjamin Nelms of Proknow for the multi-
delineator contour data presented in this work
3. AI in Radiation Oncology
3
Huynh et al. Nat Rev Clin Oncol 2020
4. 4
Netherton et al. Oncology 2021
Hype cycle for three major innovations
in radiation oncology
Automatable tasks in radiation oncology
for the modern clinic
5. Outline
• Radiomics - Decision Support Tools
- Lung Cancer Screening
- Tumor Response Prediction and Evaluation
- Aggressive Lung ADC subtype prediction
- Multimodal data: Pathology, Multiomics, etc.
• Auto Delineation and Variability Analysis
- Delineation Variability Quantification
- Dosimetric Consequences of Variabilities
- OARNet, Voxel2Mesh
5
6. Radiomics
6
¨ Controllable Feature Analysis
¨ More Interpretable
Lambin, et al. Eur J Cancer 2012
Aerts et al., Nature Communications, 2014
7. Radiomics Framework
7
Image
Registration
• Multi-level rigid
• Deformable
• Pre/Post-CT
• MSE, MI
Tumor
Segmentation
• Adaptive region growing
• Level set
• Grow cut
• Morphology filter
• Multi-modality
image segmentation
Feature
Extraction
• Intensity distribution
• Spatial variations
(texture)
• Geometric properties
• Jacobian feature
from DVF
• Feature selection
Predictive
Model
• ROC analyses
• Prediction models
• Validation
• Tumor response
• Recurrence
• Survival
Source codes: https://github.com/taznux/radiomics-tools
• Automated Workflow
- Integrate all the radiomics components
- 3D Slicer, ITK (C++), Matlab, R, and Python
- Scalable: support multicore & GPU computing
8. Radiomics Framework
8
Image
Registration
• Multi-level rigid
• Deformable
• Pre/Post-CT
• MSE, MI
Tumor
Segmentation
• Adaptive region growing
• Level set
• Grow cut
• Morphology filter
• Multi-modality
image segmentation
Feature
Extraction
• Intensity distribution
• Spatial variations
(texture)
• Geometric properties
• Jacobian feature
from DVF
• Feature selection
Predictive
Model
• ROC analyses
• Prediction models
• Validation
• Tumor response
• Recurrence
• Survival
Source codes: https://github.com/taznux/radiomics-tools
Deep Learning Model
• Automated Workflow
- Integrate all the radiomics components
- 3D Slicer, ITK (C++), Matlab, R, and Python
- Scalable: support multicore & GPU computing
9. Radiomics Framework
9
Image
Registration
• Multi-level rigid
• Deformable
• Pre/Post-CT
• MSE, MI
Tumor
Segmentation
• Adaptive region growing
• Level set
• Grow cut
• Morphology filter
• Multi-modality
image segmentation
Feature
Extraction
• Intensity distribution
• Spatial variations
(texture)
• Geometric properties
• Jacobian feature
from DVF
• Feature selection
Predictive
Model
• ROC analyses
• Prediction models
• Validation
• Tumor response
• Recurrence
• Survival
Source codes: https://github.com/taznux/radiomics-tools
Deep Learning Model
• Automated Workflow
- Integrate all the radiomics components
- 3D Slicer, ITK (C++), Matlab, R, and Python
- Scalable: support multicore & GPU computing
10. Radiomics Framework
10
Image
Registration
• Multi-level rigid
• Deformable
• Pre/Post-CT
• MSE, MI
Tumor
Segmentation
• Adaptive region growing
• Level set
• Grow cut
• Morphology filter
• Multi-modality
image segmentation
Feature
Extraction
• Intensity distribution
• Spatial variations
(texture)
• Geometric properties
• Jacobian feature
from DVF
• Feature selection
Predictive
Model
• ROC analyses
• Prediction models
• Validation
• Tumor response
• Recurrence
• Survival
Source codes: https://github.com/taznux/radiomics-tools
Deep Learning Model
• Automated Workflow
- Integrate all the radiomics components
- 3D Slicer, ITK (C++), Matlab, R, and Python
- Scalable: support multicore & GPU computing
11. Radiomics Framework
11
Image
Registration
• Multi-level rigid
• Deformable
• Pre/Post-CT
• MSE, MI
Tumor
Segmentation
• Adaptive region growing
• Level set
• Grow cut
• Morphology filter
• Multi-modality
image segmentation
Feature
Extraction
• Intensity distribution
• Spatial variations
(texture)
• Geometric properties
• Jacobian feature
from DVF
• Feature selection
Predictive
Model
• ROC analyses
• Prediction models
• Validation
• Tumor response
• Recurrence
• Survival
Source codes: https://github.com/taznux/radiomics-tools
Deep Learning Model
• Automated Workflow
- Integrate all the radiomics components
- 3D Slicer, ITK (C++), Matlab, R, and Python
- Scalable: support multicore & GPU computing
13. Lung Cancer Screening
13
¨ Early detection of lung cancer by LDCT can reduce mortality
¨ Known features correlated with PN malignancy
¤ Size, growth rate (Lung-RADS)
¤ Calcification, enhancement, solidity → texture features
¤ Boundary margins (spiculation, lobulation), attachment → shape and
appearance features
Malignant nodules Benign nodules
Size Total Malignancy
< 4mm 2038 0%
4-7 mm 1034 1%
8-20 mm 268 15%
> 20 mm 16 75%
14. ACR Lung-RADS 1.0
Category Baseline Screening Malignancy
1 No PNs; PNs with calcification
Negative
<1% chance of malignancy
2
Solid/part-solid: <6 mm
GGN: <20 mm
Benign appearance
<1% chance of malignancy
3
Solid: ≥6 to <8 mm
Part-solid: ≥6 mm with solid component <6 mm
GGN: ≥20 mm
Probably benign
1-2% chance of malignancy
4A
Solid: ≥8 to <15 mm
Part-solid: ≥8 mm with solid component ≥6 and <8 mm
Suspicious
5-15% chance of malignancy
4B
Solid: ≥15 mm
Part-solid: Solid component ≥8 mm
>15% chance of malignancy
4X
Category 3 or 4 PNs with suspicious features (e.g., enlarged lymph
nodes) or suspicious imaging findings (e.g., spiculation)
>15% chance of malignancy
14
ACR: American College of Radiology
Lung-RADS: Lung CT Screening Reporting and Data System
15. Lung Cancer Screening (Methodology)
• TCIA LIDC-IDRI public data set (n=1,010)
- Multi-institutional data
- 72 cases evaluated (31 benign and 41 malignant cases)
• Consensus contour
15
GLCM GLRM
Texture features Intensity features
2D
Shape features
3D
16. Lung Cancer Screening (SVM-LASSO Model )
16
SVM classification
Distinctive feature identification
Malignant?
Predicted malignancy
Feature extraction
Yes
10x10-fold
CV
10-fold
CV
LASSO feature selection
• Size (BB_AP) : Highly correlated with the axial longest diameter
and its perpendicular diameter (r = 0.96, larger – more
malignant)
• Texture (SD_IDM) : Tumor heterogeneity (smaller – more
malignant)
17. Lung Screening (Results: Comparison)
Sensitivity Specificity Accuracy AUC
Lung-RADS
Clinical guideline
73.3% 70.4% 72.2% 0.74
Hawkins et al. (2016)
Radiomics – 23 features
51.7 % 92.9% 80.0% 0.83
Ma et al. (2016)
Radiomics – 583 features
80.0% 85.5% 82.7%
Buty et al. (2016)
DL – 400 SH and 4096 AlexNet features
82.4%
Kumar et al. (2015)
DL: 5000 features
79.1% 76.1% 77.5%
Proposed
Radiomics: two features (Size and Texture)
87.2% 81.2% 84.6% 0.89
17
DL: Deep Learning, SH: Spherical Harmonics
Choi et al., Medical Physics, 2018.
18. ACR Lung-RADS 1.0
Category Baseline Screening Malignancy
1 No PNs; PNs with calcification
Negative
<1% chance of malignancy
2
Solid/part-solid: <6 mm
GGN: <20 mm
Benign appearance
<1% chance of malignancy
3
Solid: ≥6 to <8 mm
Part-solid: ≥6 mm with solid component <6 mm
GGN: ≥20 mm
Probably benign
1-2% chance of malignancy
4A
Solid: ≥8 to <15 mm
Part-solid: ≥8 mm with solid component ≥6 and <8 mm
Suspicious
5-15% chance of malignancy
4B
Solid: ≥15 mm
Part-solid: Solid component ≥8 mm
>15% chance of malignancy
4X
Category 3 or 4 PNs with suspicious features (e.g., enlarged lymph
nodes) or suspicious imaging findings (e.g., spiculation)
>15% chance of malignancy
18
ACR: American College of Radiology
Lung-RADS: Lung CT Screening Reporting and Data System
19. Spiculation Quantification (Motivation)
• Semantic Features
• Semi-automatic Segmentation
- GrowCut and LevelSet
19
Radiologists spiculation score (RS) for different pulmonary nodules
1 2 3 4 5
Choi et al. in CMPB 2021
24. Progression-free survival Prediction
after SBRT for early-stage NSCLC
24
Thor, Choi et al. ASTRO 2020
• 412 patients treated between 2006 and 2017
• PETs and CTs within three months prior to SBRT start.
• The median prescription dose was 50Gy in 5 fractions.
25. Progression-free survival Prediction (Results)
• PET entropy, CT number of peaks,
CT major axis, and gender.
• The most frequently selected model
included PET entropy and CT
number of peaks
- The c-index in the validation
subset was 0.77
- The prediction-stratified survival
indicated a clear separation
between the observed HR and
LR
- e.g. a PFS of 60% was observed
at 12 months in HR vs. 22
months in LR.
25
Thor, Choi et al. ASTRO 2020
26. Local tumor morphological changes
26
Jacobian Map
- Jacobian matrix: calculates rate of displacement change in each direction.
- Determinant indicates volumetric ratio of shrinkage/expansion.
012 3 = 4
012 3 > 1 volume expansion
012 3 = 1 no volume change
012 3 < 1 volume shrinkage
012 3 = 1.2 = 20% expansion
012 3 = 0.8 = 20% shrinkage (-20%)
Riyahi, Choi et al., PMB 2018
28. Local tumor morphological changes (Results)
Features P-value AUC Correlation to responders
Minimum Jacobian 0.009 0.98 -0.79
Median Jacobian 0.046 0.95 -0.72
The P-value, AUC and correlation to responders for all significant features in univariate analysis
28
Riyahi, Choi et al., PMB 2018
SVM-LASSO: AUC 0.91
29. Local Metabolic Tumor Volume Changes
29
Riyahi, Choi et al., DATRA@MICCAI 2018 AUC=0.81
30. Aggressive Lung ADC Subtype Prediction (Motivation)
30
CT
MIP
PET/CT
Soild
CT PET/CT
Five classifications of lung ADC Travis et
al. JTO 2011
¨ Solid and MIP components: poor surgery/SBRT prognosis factor
¤ Benefit from lobectomy rather than limited resection
¨ Core biopsy (Leeman et al. IJROBP 2017)
¤ Minimally invasive, not routinely performed, sampling error (about 60%
agreement with pathology)
¨ Preoperative diagnostic CT and FDG PET/CT radiomics
¤ Non-invasive and routinely performed
31. Aggressive Lung ADC Subtype Prediction (Method & Results)
• Retrospectively enrolled 120 patients
- Stage I lung ADC, ≤2cm
- Preoperative diagnostic CT and FDG PET/CT
• Histopathologic endpoint
- Aggressiveness (Solid : 18 cases, MIP : 5 cases)
• 206 radiomic features & 14 clinical parameters
• SVM-LASSO model
31
Performance of the SVM-LASSO model to predict aggressive lung ADC
Choi et al. Manuscript under review
Box plots of SUVmax (FDR q=0.004) and PET Mean of
Cluster Shade (q=0.002)
Feature Sensitivity Specificity PPV NPV Accuracy AUC
Conventional SUVmax 57.8±4.6% 78.5±1.4% 39.2±2.3% 88.6±1.1% 74.5±1.4% 0.64±0.01
SVM-LASSO PET Mean of Cluster Shade 67.4±3.1% 86.0±1.1% 53.7±2.1% 91.7±1.0% 82.4±1.0% 0.78±0.01
p-value SUVmax vs. SVM-LASSO 0.002 1e-5 7e-8 3e-5 5e-8 0.03
32. Unsupervised Learning of Deep Learned Features
from Breast Cancer Images
32
Lee, Choi et al. IEEE BIBE 2020
34. PathCNN: interpretable convolutional neural networks
for survival prediction and pathway analysis applied to glioblastoma
34
• CNNs have achieved great success
• A lack of interpretability remains a
key barrier
• Moreover, because biological array
data are generally represented in a
non-grid structured format
• PathCNN
An interpretable CNN model on
integrated multi-omics data using a
newly defined pathway image.
Oh, Choi et al. Bioinformatics 2021, joint first author Source code: https://github.com/mskspi/PathCNN.
35. PathCNN: interpretable CNNs (results)
35
Cancer PathCNN Logistic
regression SVM with RBF Neural network MiNet
GBM 0.755 ± 0.009 0.668 ± 0.039 0.685 ± 0.037 0.692 ± 0.030 0.690 ± 0.032
LGG 0.877 ± 0.007 0.816 ± 0.036 0.884 ± 0.017 0.791 ± 0.031 0.854 ± 0.027
LUAD 0.637 ± 0.014 0.581 ± 0.028 0.624 ± 0.034 0.573 ± 0.031 0.597 ± 0.042
KIRC 0.709 ± 0.009 0.654 ± 0.034 0.684 ± 0.027 0.702 ± 0.028 0.659 ± 0.030
Comparison of predictive performance with benchmark methods in terms of the area
under the curve (AUC: mean ± standard deviation) over 30 iterations of the 5-fold
cross validation
Note: AUCs for PathCNN were obtained with three principal components. Bold = Highest AUC for each dataset.
SVM, support vector machine; RBF, radial basis function; MiNet, Multi-omics Integrative Net; GBM, glioblastoma
multiforme; LGG, low-grade glioma; LUAD, lung adenocarcinoma; KIRC, kidney cancer.
Oh, Choi et al. Bioinformatics 2021, joint first author Source code: https://github.com/mskspi/PathCNN.
36. 36
A matrix of adjusted P-values. The row represents the 146 KEGG pathways ordered on pathway images.
The columns represent the first two principal components of each omics type. The red color indicates
key pathways with adjusted P-values < 0.001
37. Outline
• Radiomics - Decision Support Tools
- Lung Cancer Screening
- Tumor Response Prediction and Evaluation
- Aggressive Lung ADC subtype prediction
- Multimodal data: Pathology, Multiomics, etc.
• Auto Delineation and Variability Analysis
- Delineation Variability Quantification
- Dosimetric Consequences of Variabilities
- OARNet, Voxel2Mesh
37
38. Delineation Variability Quantification and Simulation
A framework for radiation therapy variability analysis
38
RT plan
Structure Set
CT Image
Dose Distribution
Structure Sets
DV simulation
ASSD
GrowCut
RW
Other delineators
SV analysis
DV analysis
Geometric
Dosimetric
Variability analysis
Human DV
Simulated
DV
Consensus SS
OARNet
Choi et al., AAPM, 2019.
39. Delineation Variability Quantification and
Simulation
• ESTRO Falcon contour workshop (EduCase)
- A HNC case, Larynx, 70 Gy and 35 fractions
- 14 independent manually delineated (MD) OAR structure sets (SS)
- BrainStem, Esophagus, OralCavity, Parotid_L, Parotid_R, SpinalCord,
and Thyroid
• Consensus MD SS
- The simultaneous truth and performance level estimation (STAPLE)
39
Choi, Nourzadeh et al., AAPM, 2019.
40. Delineation Variability Quantification and
Simulation (Methods)
• Geometric analysis
- Similarity: Dice coefficient (Volumetric, Surface)
- Distance: Hausdorff distance (HD), Actual Average Surface
Distance (AASD)
- Reference: STAPLE SS
• Dosimetric analysis
- Single dose distribution planned from a human SS
- DVH confidence bands (90%tile)
- !!"#$, !!#%, !!&$, !'(
40
Choi, Nourzadeh et al., AAPM, 2019.
41. Delineation Variability Quantification and Simulation (Results)
• DVH variability not predicted by geometric measures
• Large human variability
41
100%
50%
0%
100%
50%
0%
Human ASSD GrowCut RW
Right
Parotid
Left
Parotid
Choi, Nourzadeh et al., AAPM, 2019.
49. Interpretable Radiomics Toolkit
End-to-End Deep Learning Model for Malignancy Prediction
49
Input Ground Truth Voxel2Mesh
Choi et al., Manuscript in Preparation
Network AUC Accuracy Sensitivity Specificity F1
LIDC-PM Mesh Only 0.937 83.33 77.78 88.89 82.35
Mesh+Encoder 0.903 88.89 91.67 86.11 89.19
LUNGx Mesh Only 0.711 63.33 73.33 53.33 66.67
Mesh+Encoder 0.687 53.3 83.3 23.33 64.11
50. Summary
• Radiomics - Decision Support Tools
- Lung Cancer Screening
- Tumor Response Prediction and Evaluation
- Aggressive Lung ADC subtype prediction
- Multimodal data: Pathology, Multiomics, etc.
• Auto Delineation and Variability Analysis
- Delineation Variability Quantification
- Dosimetric Consequences of Variabilities
50
51. Short-term Future Works
• Develop interpretable radiomic features
- Improve spiculation quantification and multi-institution validation
- Multimodal data integration
• Human-Variability aware auto-delineation
- Variability quantification and simulation using generative models
- AI-guided interactive delineation editing
• Integrate the radiomics framework into TPS
- Eclipse (C#) and MIM (Python)
51
52. Long-term Future Works
• Comprehensive Framework for Cancer Imaging
- Multi-modal imaging
- Response prediction and evaluation (Pre, Mid, and Post)
- Longitudinal analysis of tumor change during treatment (MRgRT)
- Shape analysis (e.g., Spiculation)
- Deep learning models
• Automation of Clinical Workflow
- Big Data Analytics: EMR, PACS, ROIS, Genomics, etc.
- Provide an informatics platform for comprehensive cancer therapy
53
53. Selected Publications
1. Jung Hun Oh*, Wookjin Choi* et al., “PathCNN: interpretable convolutional neural networks for survival
prediction and pathway analysis applied to glioblastoma”, Bioinformatics, 2021, *joint first author
2. Wookjin Choi et al., “ Reproducible and Interpretable Spiculation Quantification for Lung Cancer Screening”,
Computer Methods and Programs in Biomedicine”, 2021
3. Noemi Garau, Wookjin Choi, et al., “ External validation of radiomics-based predictive models in low-dose CT
screening for early lung cancer diagnosis”, Medical Physics, 2020
4. Jiahui Wang, Wookjin Choi et al., “Prediction of anal cancer recurrence after chemoradiotherapy using
quantitative image features extracted from serial 18F-FDG PET/CT”, Frontiers in oncology, 2019
5. Wookjin Choi et al., “Radiomics analysis of pulmonary nodules in low-dose CT for early detection of lung cancer”,
Medical Physics, 2018
6. Sadegh Riyahi, Wookjin Choi, et al., “Quantifying local tumor morphological changes with Jacobian map for
prediction of pathologic tumor response to chemo-radiotherapy in locally advanced esophageal cancer”, Physics
in Medicine and Biology, 2018
7. Shan Tan, Laquan Li, Wookjin Choi, et al., “Adaptive region-growing with maximum curvature strategy for tumor
segmentation in 18F-FDG PET”, Physics in Medicine and Biology, 2017
8. Wookjin Choi et al., “Individually Optimized Contrast-Enhanced 4D-CT for Radiotherapy Simulation in Pancreatic
Ductal Adenocarcinoma”, Medical Physics, 2016
9. Wookjin Choi, Tae-Sun Choi, “Automated Pulmonary Nodule Detection based on Three-dimensional Shape-based
Feature Descriptor”, Computer Methods and Programs in Biomedicine, 2014
54
Complete list of publications: https://scholar.google.com/citations?user=iHgsGLUAAAAJ
54. Post-Doctoral Research Fellow
Developing Interpretable Predictive Models for Radiation
Therapy
• PI: Wookjin Choi, PhD - Wookjin.Choi@Jefferson.edu
• 2 Years
• Machine Learning/Deep Learning: Radiomics (PET/CT & MR) and Bioinformatics
• Computational Medical Physics: Development of Predictive Models and
Automated Workflows, and Improve Clinical Workflow
• Internal or Extramural Research Funding Opportunities
Qualifications
• Ph.D. in Computer Science, Electrical Engineering, Medical Physics, or related
field required