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.

1. AI to Reduce Radiation-induced Cardiotoxicity in Lung Cancer Radiotherapy:
A Novel Functional Radiomics
using Cardiac FDG-PET/CT
Wookjin Choi, Adam P. Dicker, and Yevgeniy Vinogradskiy
Department of Radiation Oncology,
Sidney Kimmel Medical College / Cancer Center
at Thomas Jefferson University, Philadelphia, PA

2. • Assistant Professor / Computational Medical Physicist
Sidney Kimmel Medical College / Cancer Center at Thomas Jefferson University
• A research grant from ViewRay, Inc.
• A research grant from Varian Medical Systems, Inc.
• A research grant from the PROPEL Center
• Research software license for INT Contour from Carina Medical
• This project was partially supported by the Sidney Kimmel Cancer Center Support Grant
(NIH/NCI P30 CA056036)
2
Disclosure

3. • Introduction
Radiation Oncology
Medical Physics
Radiomics
• Heart Toxicity Prediction
Method
Results
• Conclusion
3
Overview

4. 4
Cancer Therapy

5. • Radiation has been an effective tool for
treating cancer for more than 100 years
• More than 60 percent of patients diagnosed
with cancer will receive radiation therapy as
part of their treatment
• Radiation oncologists are cancer specialists
who manage the care of cancer patients with
radiation for either cure or palliation
5
Radiation Oncology
Patient being treated with modern
radiation therapy equipment.

6. 6
Advances in Radiation Oncology
Figure 1(b) of Lee et al. "Physical and radiobiological evaluation of radiotherapy treatment plan." Evolution of ionizing radiation research. InTech (2015): 109-50.

7. • Radiation Oncologist prescribes and oversees the radiation therapy treatments
• Medical Physicist ensures that treatment plans are properly tailored for each
patient, and is responsible for the calibration and accuracy of treatment equipment
• Dosimetrist works with the radiation oncologist and medical physicist to calculate
the proper dose of radiation given to the tumor
• Radiation Therapist administers the daily radiation under the doctor’s prescription
and supervision
• Radiation Oncology Nurse interacts with the patient and family at the time of
consultation, throughout the treatment process and during follow-up care
7
The Radiation Oncology Team

8. • Medical physics is easy to define
 Application of physics to medicine
• There is a long tradition of using applied physics to improve medicine (both
treatments and diagnostics)
8
Medical Physics
From the early radiographs to the modern medical linear accelerator

9. 9
AI in Radiation Oncology
Huynh et al. Nat Rev Clin Oncol 2020
A general overview of the radiation therapy workflow with brief descriptions of expected applications of AI at each step.

10. Diagnosis Simulation Contouring
Treatment
Planning
Treatment
QA
Treatment
Delivery
MIM Protégé AI &
Auto export to
Eclipse
Auto-contouring
of daily
CBCT/MR
Eclipse
DVH Metric
script
MIM Archive &
Auto export to
RadCalc
AI Dose
Prediction using
Diagnostic
PET/CT
Weekly
Dashboard
Dosimetry
Whiteboard
Longitudinal
Radiomics
Customized AI
Contouring
Current Solutions in Jefferson SKCC Radiation Oncology
10
Cardiac PET
Heart Toxicity
Prediction
Cardiac PET
Heart Toxicity
Assessment

11. Radiomics
11
Aerts et al. Nature Communications, 2014
Figure 1: Extracting radiomics data from images. Figure 2: Analysis workflow.
Decoding tumour phenotype by noninvasive imaging using a quantitative radiomics approach

12. • Quantitative approach to medical imaging
• Extracts many features from medical images
• These features can be used to predict prognosis and therapeutic response
• Radiomics has the potential to revolutionize the way we diagnose and treat cancer
12
Radiomics
Fully automated pipeline, Woznicki et al. Frontiers in Radiology 2022 https://github.com/pwoznicki/AutoRadiomics

13. A Novel
Functional
Radiomics
using Cardiac
FDG-PET/CT

14. • Traditional methods of evaluating cardiotoxicity:
 Focus solely on radiation doses to the heart.
 Do not incorporate functional imaging information.
• Functional imaging has potential to improve cardiotoxicity prediction for
lung cancer patients undergoing radiotherapy:
 FDG PET/CT imaging is routinely obtained for lung cancer staging.
 PET scans can be used to evaluate cardiac inflammation.
 Studies indicate that PET cardiac signal can predict clinical outcomes.
• The purpose of this work:
 Develop a radiomics model to predict clinical cardiac assessment using FDG PET/CT
scans.
Heart Toxicity Prediction using PET/CT Radiomics for Lung Radiotherapy
14

15. • Pre-Treatment only - before treatment
 Dose prediction
 Cardiac PET/CT Radiomics to predict existing heart disease
 CT coronary artery calcium scoring
 Collected TJU 100 pts data and continue to collect more data
 70 pts from ACRIN 6668/RTOG 0235, 39 pts from CU/BU
 Manual and auto heart contouring
• Pre & Post Treatment – during/after treatment
 PET/CT Radiomics
 Delta-radiomics between pre and post PET/CT
 70 pts from ACRIN 6668/RTOG 0235, 39 pts from CU/BU
 Manual and auto heart contouring
15
Heart Toxicity Prediction using PET/CT Radiomics for Lung Radiotherapy

16. 16
AI in Radiation Oncology – Heart Toxicity Prediction
Huynh et al. Nat Rev Clin Oncol 2020
A general overview of the radiation therapy workflow with brief descriptions of expected applications of AI at each step.
Dose Map
Prediction
Cardiac Uptake
Classification
Overall Survival
Prediction

17. Diagnostic PET/CT Radiation Target delineation Normal tissue delineation Radiation dose distribution
Diagnostic PET/CT Radiation dose distribution
AI/ML
Dose Prediction before Treatment Planning
17

18. • The study included 209 pre-treatment 18F-FDG PET-CT images of lung cancer patients from 3
different study populations (TJU: 100, ACRIN: 70, CU: 39).
 Low quality images were excluded 7 (TJU: 1, ACRIN: 4, CU:2).
Cardiac PET/CT Radiomics - Dataset
CLASS
LV MYOCARDIAL
UPTAKE PATTERN
TJU ACRIN CU TOTAL
0 Uniform 19 (19.2%) 21 (31.8%) 11 (29.8%) 51 (25.2%)
1 Absent 44 (44.4%) 23 (34.8%) 13 (35.1%) 80 (39.6%)
2 Heterogeneous 36 (36.4%) 22 (33.3%) 13 (35.1%) 71 (35.1%)
ALL 99 (100.0%) 66 (100.0%) 37 (100.0%) 202 (100.0%)
Table 1. Statistics of LV FDG Uptake Pattern Classification
Figure 2. Cardiac uptake classifications, Uniform, Absent, and Heterogeneous (Non-uniform and Focal )
18

19. 19
Method - Study Flow Diagram
Figure 1. Study and cardiac model diagram, training dataset: TJU (N=99), and external validation data sets: ACRIN (N=66) and CU (N=37).
RFE: Recursive Feature Elimination, SFS: Sequential Feature Selection
1. Pre-processing
LV FDG uptake patter
review
By Radiation Oncologist
Manual
Heart Contouring
By Medical Physicist
Automatic
Heart Contouring
By 1Commercial Software
Training
Data set
(N=99)
2. Feature Analysis
Feature Extraction
- PyRadiomics
- In-house SW
Feature Reduction
- Wilcoxon test
- Clustering
Feature Selection
- RFE
- SFS
3. Model Building
Model Exploration
- TPOT
Optimization
- Hyper-parameters
- Feature Selection
Evaluation
- Validation
- External validation
Predicted
LV Uptake
Pattern
External
Validation
Data sets
(N=103)
1INTContour, Carina Medical https://www.carinaai.com/intcontour.html

20. • Training and Validation Data:
 Training dataset: TJU (N=100), 80% training set (TPOT optimization), 20% validation set.
 External validation datasets: ACRIN (N=70), CU (N=39)
• Radiomics Feature Extraction
 Heart delineation: Manual and Auto
 Selected 200 novel functional radiomics features – In-house: 103, PyRadiomics: 97
 Feature categories include Shape (2D and 3D), First-order (2D and 3D), GLCM, GLSZM, GLRLM, NGTDM,
GLDM.
• Feature Reduction
 Feature robustness test using ICC between features from Auto and Manual delineations
 Wilcoxon test (Bonferroni adjusted p<0.05)
 Hierarchical clustering
Method – Radiomics Feature Extraction and Reduction
20

21. 21
Results: Pipeline Optimization using TPOT
Feature Selector
Variance Threshold
Threshold: 0.0001
56 features
Threshold: 0.005
52 features
RFE ranking and
SFS with TPOT optimization
Top 9 features
Feature Transformer
One Hot Encoder
Minimum_fraction: 0.15,
Sparse: False,
Threshold: 10
Minimum_fraction: 0.25
Minimum_fraction: 0.15
Classifier
Extra Trees Classifier
Bootstrap: False,
criterion: gini,
max_features: 0.95,
min_samples_leaf: 3,
min_samples_split: 20
Criterion: entropy,
max_features: 0.75,
min_samples_leaf: 2,
min_samples_split: 17
Bootstrap: True,
criterion: gini,
max_features: 0.8,
min_samples_split: 10
Model
Discovery
Hyper-
parameter
Optimization
Feature
Optimization
Validation Accuracy (%)
TJU ACRIN CU
90.0 78.8 90.2
90.0 78.8 91.9
95.0 80.3 91.9

22. • TPOT (Tree-Based Pipeline Optimization Tool)
 100 populations, 10 generations, 10-fold CV
 Evaluated 10,000 pipeline configurations to identify the best model.
• Pipeline Construction
 Three component template: Feature Selector—Feature Transformer—Classifier
o Feature Selector: P-value (ANOVA F-statistic), Family Wise Error Rate, Percentile, Low-Variance Feature Removal, RFE
with Extra Trees Classifier
o Feature Transformer: Binarization, ICA, Feature Agglomeration (Various Linkage & Affinity Methods), Scaling (Max
Absolute Value, Min-Max Values), Normalization (l1, l2, max norm) ,Kernel Approximation (RBF, Cosine, χ2, etc.), PCA,
Polynomial Features, RBF Sampler, Robust Scaler, Standard Scaler, Zero Count, One Hot Encoder
o Classifier: Naïve Bayesian (Bernoulli, Multinomial), Decision Tree Classifier, Extra Trees Classifier, Random Forest
Classifier, Gradient Boosting Classifier, XGBoost, KNN Classifier, Linear SVM Classifier, Logistic Regression, SGD, MLP
• Hyper-parameter Optimization: optimize parameters of the discovered pipeline
• Feature Optimization: Utilized RFE for feature ranking followed by SFS with TPOT optimization.
Method - Machine Learning Model Pipeline Selection and Optimization
22

23. Results – Feature reduction and Prediction results
Model iteration and explanation
Number of
features
10-fold cross
validation
accuracy (%)
Training
accuracy (%)
Validation
accuracy (%)
External
Validation
accuracy -
ACRIN (%)
External
Validation
accuracy – CU
(%)
Iteration 1:
Model discovery 56 88.9 94.9 90.0 78.8 89.2
Iteration 2:
Hyperparameter optimization
52 89.9 96.2 90.0 78.8 91.9
Iteration 3:
Feature optimization 9 89.8 92.4 95.0 80.3 91.9
 9 Selected Features: 2D Maximum, 2D Skewness, Mean, GLCM Cluster Prominence, GLCM Sum
Average, GLCM Cluster Shade, GLDM Dependence Variance, Maximum, 2D Kurtosis
Feature reduction The number of remaining Features
ICC 153
Wilcoxon Test 141
Hierarchical Clustering 62
Table 3. Model pipeline optimization results and validation results.
Table 2. Feature reduction results
23

24. • Delta-Radiomics calculation:
The change between pre- and post-PET/CT.
 Difference Δ𝐹𝑖
= 𝐹𝑝𝑜𝑠𝑡
𝑖
− 𝐹𝑝𝑟𝑒
𝑖
𝐹: the feature vector, 𝑖: the index of the feature vector
• The study identified four clinically relevant delta-
radiomics features
 Pre-CT Maximum
 Post-CT Minimum
 Delta-CT GLRM Run Variance
 Delta-PET GLRM Run Entropy
• Model performance
 CV on the training set: 0.91 AUC
 The test set: 0.87 AUC
24
Results - Overall survival prediction using delta radiomics Pre & Post
Figure 1: Example of patients with variable pre and post-treatment
cardiac PET/CT signal.
Vinogradskiy Y, Diot Q, Jones B, et al. Evaluating Positron Emission Tomography-
Based Functional Imaging Changes in the Heart After Chemo-Radiation for Patients
With Lung Cancer. International Journal of Radiation Oncology* Biology* Physics.
2020;106(5):1063-1070.

25. • Utilized dataset of 209 patients for functional cardiac radiomics model
development.
• Achieved high accuracy:
 9 optimal features were selected
 Training set validation: TJU: 95%
 External validation : ACRIN: 80.3%, CU: 91.1%
• Clinical Application:
 Automated prediction of existing cardiac conditions.
 Early biomarker for post-radiotherapy cardiac complications.
Conclusion
25

26. Thank you !

27. Jefferson.edu