SlideShare a Scribd company logo
Presenter: Anjani K. Dhrangadhariya
Joint work with: G. Aguilar, T. Solorio, R. Hilfiker, H. Müller
CLEF 2021, Bucharest, Romania, 23.09.2021
1
End-to-end Fine-grained Neural
Entity Recognition of Patients,
Interventions, Outcomes
Introduction
2
…A semi-structured interview
was used to obtain qualitative
information on the effect of the
aerobics intervention vs.
conventional exercise. The
convenience sample included 15
adult Oncology outpatients, 13
female and 2 male, ranging in
age from 20 to 87. Quality of
life, fatigue, and side-effects
were the measured outcomes…
Participant
Intervention
Comparator
Outcome
Introduction
3
…A semi-structured interview
was used to obtain qualitative
information on the effect of the
aerobics intervention vs.
conventional exercise. The
convenience sample included 15
adult Oncology outpatients, 13
female and 2 male, ranging in
age from 20 to 87. Quality of
life, fatigue, and side-effects
were the measured outcomes…
P
I
C
O
aerobics intervention
Quality of life
fatigue
conventional
exercise
15
Adult, age from 20 to 87
Oncology outpatients
13 female and 2 male
side-effects
Sample size
Age
Condition
Gender
Physical
Control
Physical
Physical
Adverse
effects
Coarse-grained
labels
Fine-grained
labels
4
P
I
C
O
aerobics intervention
Quality of life
fatigue
conventional
exercise
15
Adult, age from 20 to 87
Oncology outpatients
13 female and 2 male
side-effects
Sample size
Age
Condition
Gender
Physical
Control
Physical
Physical
Adverse
effects
Systematic Reviews (SR)
Inclusion into the SR Exclude from the SR
• Evidence-based clinical practice
• Health policy-making
Introduction
Challenges
• Coarse-grained PICO - Disagreement between human
annotators
• Fine-grained PICO subtypes - are neither clearly identified
nor standardized as semantic units.
• Domain-specific challenges
• Naming conventions are not established in the health vs. pharma
literature
5
Brockmeier, Austin J., et al. "Improving reference prioritisation with PICO recognition." BMC medical informatics and decision making 19.1
(2019): 1-14.
Challenges
• Sentence-level PICO extraction
• Only available “large” entity annotation PICO dataset – EBM-PICO
• ~5000 PICO annotated RCT abstract
• Manually annotated – PICO and subtypes
• Training and Test sets
• Annotators: Medical students and physicians
• Only available with multi-level annotation
6
Nye, Benjamin, et al. "A corpus with multi-level annotations of patients, interventions and outcomes to support language processing for medical
literature." Proceedings of the conference. Association for Computational Linguistics. Meeting. Vol. 2018. NIH Public Access, 2018.
State of the art
Nye et al. Hand-engineered features
Poor performance on the fine-grained PICO
Ananiadou et al. Do not focus on the fine-grained recognition
Beltagy et al. SciBERT domain-adaptation improved coarse-grained recognition.
Zhang et al. Disease NER - Participant condition entity only
Xu et al. Focus on patient demographics (gender, sex, ethnicity) alone
Chung et al. Focus on the Intervention arms in the RCTs
7
1. Most studies concentrate on the coarse-grained PICO recognition
but not on the extraction of PICO subtypes and are not end-to-end.
• PICO recognition = Semi-automation
• PICO subtype recognition = full automation
2. No focus on domain differences
3. No exploration of using coarse and fine-grained labels annotation
together – Multi-task learning (MTL)?
8
Motivation
Methodology: MTL
Single-task learning Multi-task learning
• Task A, Task B
• Labeled dataset d(xa ,ya), d(xb ,yb)
• Loss function La, Lb
• ModelA, ModelB
• Task A, Task B
• Labeled dataset d(x,ya), d(x,yb)
• A loss function L
• Model
Models trained separately and do not
share the L and the internal parameters
Models trained in unison and share the L
and the internal parameters
9
• Train one neural network to do multiple related tasks where each task hopefully
helps improve the performance on the other task.
https://ruder.io/multi-task/index.html
Methodology: MTL
• The key idea is that one task (auxiliary task) can help another task (main task).
• Assumption: Fine-grained entities nested under the coarse-grained entities =
related tasks.
• Mutual sources of inductive bias
10
Coarse-grained Participant labels Fine-grained Participant labels
Methodology: Datasets
12
• EBM-PICO training set
• 5000 PICO annotated RCT abstracts
• PICO coarse- and fine-grained
• Training set: 4,993 documents
• Annotated by medical students
EBM-PICO evaluation corpus Hilfiker corpus
• 191 PICO documents (fine and coarse)
• Annotated by physicians
• Pharma domain
• 153 PICO annotated (fine and coarse)
• Annotated by the first author
• Physiotherapy and Rehab domain
https://www.ncbi.nlm.nih.gov/pmc/articles/PMC6174533/bin/NIHMS988059-supplement-Appendix.pdf
Documents = RCT abstracts
13
Prepares the input text
tokens by subword
tokenizing them using
BERT tokenizer
Methodology: MTL architecture
14
Methodology: MTL architecture
Used to extract deep
contextualized BERT
vectors from the
tokenized text
15
Methodology: MTL architecture
Average over the last
four BERT layers are
calculated and the
representation is fed to
the BiLSTM layer
16
Methodology: MTL architecture
BiLSTM outputs are
concatenated and self-
attention is applied
17
Methodology: MTL architecture
Emission
probabilities from
the self-attention
layer are fed into
the two separate
CRF heads along
with the true
labels
18
Methodology: MTL architecture
=
Decoder layers
separately learn
the coarse- and
fine- grained
labels on the
data and
calculate
individual losses
1. MTL
2. STL
Experiments conducted
19
Experiment Model setup
I BERT Linear
II BERT LSTM CRF
III BERT BiLSTM CRF
IV BERT LSTM Atten CRF
V BERT BiLSTM Atten CRF
V BERT BiLSTM Multihead Atten CRF
VI BERT BiLSTM Multihead Atten CRF (fine-)/Linear(coarse-)
Experiments conducted
20
Experiment Model setup
I BERT Linear
II BERT LSTM CRF
III BERT BiLSTM CRF
IV BERT LSTM Atten CRF
V BERT BiLSTM Atten CRF
Ablation 1 BERT BiLSTM attn (on coarse) CRF
Ablation 2 BERT BiLSTM attn (on fine) CRF
V BERT BiLSTM Multihead Atten CRF
VI BERT BiLSTM Multihead Atten CRF (fine-)/Linear(coarse-)
1. MTL
2. STL
Results
21
Underline = Significant improvement compared to the counterpart
MTL performs better
(BOLD) for some
experimental setups
but it is not a
significant
improvement
(UNDERLINE)
Results: Ablation experiments
22
Error analysis (EBM-PICO eval set)
23
• Intervention entity : Poor performance
• “Psychological” and “Other” classes
• Fewer annotations
• Consistently confused between “Educational” and
“Psychological” – Annotation exercise
• Plenty PICO entities get classified as out-of-the-
span label and vice versa.
• Class overlap + Class imbalance
Conclusion
1. Propose two end-to-end neural model setups for fine-grained PICO
recognition.
2. Both model setups outperform the previous approaches for the
fine-grained ``Participant'' and ``Outcome'' entity recognition.
3. MTL is a good alternative to STL but not directly.
4. We contribute a PICO annotated corpus for the physiotherapy
domain and explore the domain difference
• Our error analysis warrants rethinking of semantically solid class
definitions for fine-grained PICO entities along with ontology
development for the healthcare domain.
24
Training setup
1. Epochs 15
2. Max. seq. length BERT 512
3. BERTs’ Last four layers summed
4. BERT not freezed
5. LSTM/BiLSTM hidden size – 512/1025
6. Optimizer – AdamW, Learning rate 5e-5
7. Gradients clipped to one.
8. Weighted cross-entropy loss using Sklearn suite
9. A single Tesla K80 GPU, PyTorch
25
References
• Boudin, Florian, et al. "Combining classifiers for robust PICO element detection." BMC
medical informatics and decision making 10.1 (2010): 1-6.
• Huang, Ke-Chun, et al. "Classification of PICO elements by text features systematically
extracted from PubMed abstracts." 2011 IEEE International Conference on Granular
Computing. IEEE, 2011.
• Jin, Di, and Peter Szolovits. "Pico element detection in medical text via long short-term
memory neural networks." Proceedings of the BioNLP 2018 workshop. 2018.
• Beltagy, Iz, Kyle Lo, and Arman Cohan. "Scibert: A pretrained language model for
scientific text." arXiv preprint arXiv:1903.10676 (2019).
• Zhang, Tengteng, et al. "Unlocking the power of deep pico extraction: Step-wise medical
ner identification." arXiv preprint arXiv:2005.06601 (2020).
• Xu, Rong, et al. "Extracting subject demographic information from abstracts of
randomized clinical trial reports." Medinfo 2007: Proceedings of the 12th World Congress
on Health (Medical) Informatics; Building Sustainable Health Systems. IOS Press, 2007.
26
Thank you for your attention
Questions?
Anjani Dhrangadhariya
anjani.dhrangadhariya@hevs.ch
https://www.linkedin.com/in/anjani-dhrangadhariya/
More information
http://medgift.hevs.ch/wordpress/
https://www.examode.eu/
27

More Related Content

What's hot

IMAGE SEGMENTATION USING FCM ALGORITM | J4RV3I12021
IMAGE SEGMENTATION USING FCM ALGORITM | J4RV3I12021IMAGE SEGMENTATION USING FCM ALGORITM | J4RV3I12021
IMAGE SEGMENTATION USING FCM ALGORITM | J4RV3I12021
Journal For Research
 
International Journal of Biometrics and Bioinformatics(IJBB) Volume (3) Issue...
International Journal of Biometrics and Bioinformatics(IJBB) Volume (3) Issue...International Journal of Biometrics and Bioinformatics(IJBB) Volume (3) Issue...
International Journal of Biometrics and Bioinformatics(IJBB) Volume (3) Issue...CSCJournals
 
IRJET- Breast Cancer Detection from Histopathology Images: A Review
IRJET-  	  Breast Cancer Detection from Histopathology Images: A ReviewIRJET-  	  Breast Cancer Detection from Histopathology Images: A Review
IRJET- Breast Cancer Detection from Histopathology Images: A Review
IRJET Journal
 
IRJET- Analysis of Brain Tumor Classification by using Multiple Clustering Al...
IRJET- Analysis of Brain Tumor Classification by using Multiple Clustering Al...IRJET- Analysis of Brain Tumor Classification by using Multiple Clustering Al...
IRJET- Analysis of Brain Tumor Classification by using Multiple Clustering Al...
IRJET Journal
 
IRJET - Detection and Classification of Brain Tumor
IRJET - Detection and Classification of Brain TumorIRJET - Detection and Classification of Brain Tumor
IRJET - Detection and Classification of Brain Tumor
IRJET Journal
 
IRJET- Brain Tumor Segmentation and Detection using F-Transform
IRJET- Brain Tumor Segmentation and Detection using F-TransformIRJET- Brain Tumor Segmentation and Detection using F-Transform
IRJET- Brain Tumor Segmentation and Detection using F-Transform
IRJET Journal
 
Detection of erythemato-squamous diseases using AR-CatfishBPSO-KSVM
Detection of erythemato-squamous diseases using AR-CatfishBPSO-KSVMDetection of erythemato-squamous diseases using AR-CatfishBPSO-KSVM
Detection of erythemato-squamous diseases using AR-CatfishBPSO-KSVM
sipij
 
2015 bioinformatics wim_vancriekinge
2015 bioinformatics wim_vancriekinge2015 bioinformatics wim_vancriekinge
2015 bioinformatics wim_vancriekinge
Prof. Wim Van Criekinge
 
Detecting malaria using a deep convolutional neural network
Detecting malaria using a deep  convolutional neural networkDetecting malaria using a deep  convolutional neural network
Detecting malaria using a deep convolutional neural network
Yusuf Brima
 
Comparing prediction accuracy for machine learning and
Comparing prediction accuracy for machine learning andComparing prediction accuracy for machine learning and
Comparing prediction accuracy for machine learning andAlexander Decker
 
CLASSIFICATION OF CANCER BY GENE EXPRESSION USING NEURAL NETWORK
CLASSIFICATION OF CANCER BY GENE EXPRESSION USING NEURAL NETWORKCLASSIFICATION OF CANCER BY GENE EXPRESSION USING NEURAL NETWORK
CLASSIFICATION OF CANCER BY GENE EXPRESSION USING NEURAL NETWORK
International Research Journal of Modernization in Engineering Technology and Science
 
A Wavelet Based Automatic Segmentation of Brain Tumor in CT Images Using Opti...
A Wavelet Based Automatic Segmentation of Brain Tumor in CT Images Using Opti...A Wavelet Based Automatic Segmentation of Brain Tumor in CT Images Using Opti...
A Wavelet Based Automatic Segmentation of Brain Tumor in CT Images Using Opti...
CSCJournals
 
MELANOMA CELL DETECTION IN LYMPH NODES HISTOPATHOLOGICAL IMAGES USING DEEP LE...
MELANOMA CELL DETECTION IN LYMPH NODES HISTOPATHOLOGICAL IMAGES USING DEEP LE...MELANOMA CELL DETECTION IN LYMPH NODES HISTOPATHOLOGICAL IMAGES USING DEEP LE...
MELANOMA CELL DETECTION IN LYMPH NODES HISTOPATHOLOGICAL IMAGES USING DEEP LE...
sipij
 
ISMB2014読み会 イントロ + Deep learning of the tissue-regulated splicing code
ISMB2014読み会 イントロ + Deep learning of the tissue-regulated splicing codeISMB2014読み会 イントロ + Deep learning of the tissue-regulated splicing code
ISMB2014読み会 イントロ + Deep learning of the tissue-regulated splicing code
Kengo Sato
 
Skin Cancer Detection using Image Processing in Real Time
Skin Cancer Detection using Image Processing in Real TimeSkin Cancer Detection using Image Processing in Real Time
Skin Cancer Detection using Image Processing in Real Time
ijtsrd
 
IRJET- A Study on Brain Tumor Detection Algorithms for MRI Images
IRJET- A Study on Brain Tumor Detection Algorithms for MRI ImagesIRJET- A Study on Brain Tumor Detection Algorithms for MRI Images
IRJET- A Study on Brain Tumor Detection Algorithms for MRI Images
IRJET Journal
 
IRJET - Lung Disease Prediction using Image Processing and CNN Algorithm
IRJET -  	  Lung Disease Prediction using Image Processing and CNN AlgorithmIRJET -  	  Lung Disease Prediction using Image Processing and CNN Algorithm
IRJET - Lung Disease Prediction using Image Processing and CNN Algorithm
IRJET Journal
 
BRAIN TUMOR CLASSIFICATION IN 3D-MRI USING FEATURES FROM RADIOMICS AND 3D-CNN...
BRAIN TUMOR CLASSIFICATION IN 3D-MRI USING FEATURES FROM RADIOMICS AND 3D-CNN...BRAIN TUMOR CLASSIFICATION IN 3D-MRI USING FEATURES FROM RADIOMICS AND 3D-CNN...
BRAIN TUMOR CLASSIFICATION IN 3D-MRI USING FEATURES FROM RADIOMICS AND 3D-CNN...
IAEME Publication
 

What's hot (18)

IMAGE SEGMENTATION USING FCM ALGORITM | J4RV3I12021
IMAGE SEGMENTATION USING FCM ALGORITM | J4RV3I12021IMAGE SEGMENTATION USING FCM ALGORITM | J4RV3I12021
IMAGE SEGMENTATION USING FCM ALGORITM | J4RV3I12021
 
International Journal of Biometrics and Bioinformatics(IJBB) Volume (3) Issue...
International Journal of Biometrics and Bioinformatics(IJBB) Volume (3) Issue...International Journal of Biometrics and Bioinformatics(IJBB) Volume (3) Issue...
International Journal of Biometrics and Bioinformatics(IJBB) Volume (3) Issue...
 
IRJET- Breast Cancer Detection from Histopathology Images: A Review
IRJET-  	  Breast Cancer Detection from Histopathology Images: A ReviewIRJET-  	  Breast Cancer Detection from Histopathology Images: A Review
IRJET- Breast Cancer Detection from Histopathology Images: A Review
 
IRJET- Analysis of Brain Tumor Classification by using Multiple Clustering Al...
IRJET- Analysis of Brain Tumor Classification by using Multiple Clustering Al...IRJET- Analysis of Brain Tumor Classification by using Multiple Clustering Al...
IRJET- Analysis of Brain Tumor Classification by using Multiple Clustering Al...
 
IRJET - Detection and Classification of Brain Tumor
IRJET - Detection and Classification of Brain TumorIRJET - Detection and Classification of Brain Tumor
IRJET - Detection and Classification of Brain Tumor
 
IRJET- Brain Tumor Segmentation and Detection using F-Transform
IRJET- Brain Tumor Segmentation and Detection using F-TransformIRJET- Brain Tumor Segmentation and Detection using F-Transform
IRJET- Brain Tumor Segmentation and Detection using F-Transform
 
Detection of erythemato-squamous diseases using AR-CatfishBPSO-KSVM
Detection of erythemato-squamous diseases using AR-CatfishBPSO-KSVMDetection of erythemato-squamous diseases using AR-CatfishBPSO-KSVM
Detection of erythemato-squamous diseases using AR-CatfishBPSO-KSVM
 
2015 bioinformatics wim_vancriekinge
2015 bioinformatics wim_vancriekinge2015 bioinformatics wim_vancriekinge
2015 bioinformatics wim_vancriekinge
 
Detecting malaria using a deep convolutional neural network
Detecting malaria using a deep  convolutional neural networkDetecting malaria using a deep  convolutional neural network
Detecting malaria using a deep convolutional neural network
 
Comparing prediction accuracy for machine learning and
Comparing prediction accuracy for machine learning andComparing prediction accuracy for machine learning and
Comparing prediction accuracy for machine learning and
 
CLASSIFICATION OF CANCER BY GENE EXPRESSION USING NEURAL NETWORK
CLASSIFICATION OF CANCER BY GENE EXPRESSION USING NEURAL NETWORKCLASSIFICATION OF CANCER BY GENE EXPRESSION USING NEURAL NETWORK
CLASSIFICATION OF CANCER BY GENE EXPRESSION USING NEURAL NETWORK
 
A Wavelet Based Automatic Segmentation of Brain Tumor in CT Images Using Opti...
A Wavelet Based Automatic Segmentation of Brain Tumor in CT Images Using Opti...A Wavelet Based Automatic Segmentation of Brain Tumor in CT Images Using Opti...
A Wavelet Based Automatic Segmentation of Brain Tumor in CT Images Using Opti...
 
MELANOMA CELL DETECTION IN LYMPH NODES HISTOPATHOLOGICAL IMAGES USING DEEP LE...
MELANOMA CELL DETECTION IN LYMPH NODES HISTOPATHOLOGICAL IMAGES USING DEEP LE...MELANOMA CELL DETECTION IN LYMPH NODES HISTOPATHOLOGICAL IMAGES USING DEEP LE...
MELANOMA CELL DETECTION IN LYMPH NODES HISTOPATHOLOGICAL IMAGES USING DEEP LE...
 
ISMB2014読み会 イントロ + Deep learning of the tissue-regulated splicing code
ISMB2014読み会 イントロ + Deep learning of the tissue-regulated splicing codeISMB2014読み会 イントロ + Deep learning of the tissue-regulated splicing code
ISMB2014読み会 イントロ + Deep learning of the tissue-regulated splicing code
 
Skin Cancer Detection using Image Processing in Real Time
Skin Cancer Detection using Image Processing in Real TimeSkin Cancer Detection using Image Processing in Real Time
Skin Cancer Detection using Image Processing in Real Time
 
IRJET- A Study on Brain Tumor Detection Algorithms for MRI Images
IRJET- A Study on Brain Tumor Detection Algorithms for MRI ImagesIRJET- A Study on Brain Tumor Detection Algorithms for MRI Images
IRJET- A Study on Brain Tumor Detection Algorithms for MRI Images
 
IRJET - Lung Disease Prediction using Image Processing and CNN Algorithm
IRJET -  	  Lung Disease Prediction using Image Processing and CNN AlgorithmIRJET -  	  Lung Disease Prediction using Image Processing and CNN Algorithm
IRJET - Lung Disease Prediction using Image Processing and CNN Algorithm
 
BRAIN TUMOR CLASSIFICATION IN 3D-MRI USING FEATURES FROM RADIOMICS AND 3D-CNN...
BRAIN TUMOR CLASSIFICATION IN 3D-MRI USING FEATURES FROM RADIOMICS AND 3D-CNN...BRAIN TUMOR CLASSIFICATION IN 3D-MRI USING FEATURES FROM RADIOMICS AND 3D-CNN...
BRAIN TUMOR CLASSIFICATION IN 3D-MRI USING FEATURES FROM RADIOMICS AND 3D-CNN...
 

Similar to End-to-end Fine-grained Neural Entity Recognition of Patients, Interventions, Outcomes

DISTANT-CTO: A Zero Cost, Distantly Supervised Approach to Improve Low-Resour...
DISTANT-CTO: A Zero Cost, Distantly Supervised Approach to Improve Low-Resour...DISTANT-CTO: A Zero Cost, Distantly Supervised Approach to Improve Low-Resour...
DISTANT-CTO: A Zero Cost, Distantly Supervised Approach to Improve Low-Resour...
Anjani Dhrangadhariya
 
Conferencia del Dr. Ander Ramos en #Perspectives2015
Conferencia del Dr. Ander Ramos en #Perspectives2015Conferencia del Dr. Ander Ramos en #Perspectives2015
Conferencia del Dr. Ander Ramos en #Perspectives2015
TECNALIA Research & Innovation
 
EUSFLAT 2019: explainable neuro fuzzy recurrent neural network to predict col...
EUSFLAT 2019: explainable neuro fuzzy recurrent neural network to predict col...EUSFLAT 2019: explainable neuro fuzzy recurrent neural network to predict col...
EUSFLAT 2019: explainable neuro fuzzy recurrent neural network to predict col...
Servio Fernando Lima Reina
 
A review on early hospital mortality prediction using vital signals
A review on early hospital mortality prediction using vital signalsA review on early hospital mortality prediction using vital signals
A review on early hospital mortality prediction using vital signals
Reza Sadeghi
 
SNOMED CT concept model for molecular pathology_final.pptx
SNOMED CT concept model for molecular pathology_final.pptxSNOMED CT concept model for molecular pathology_final.pptx
SNOMED CT concept model for molecular pathology_final.pptx
HariHaran685388
 
Weakly supervised PICO information extraction using Snorkel
Weakly supervised PICO information extraction using SnorkelWeakly supervised PICO information extraction using Snorkel
Weakly supervised PICO information extraction using Snorkel
Anjani Dhrangadhariya
 
Real-time model for adaptive radiation therapy a biomechanical approach using...
Real-time model for adaptive radiation therapy a biomechanical approach using...Real-time model for adaptive radiation therapy a biomechanical approach using...
Real-time model for adaptive radiation therapy a biomechanical approach using...
Az.Ospedaliero-Universitaria di Modena
 
A new generic approach for scoping HTA. Iris Pasternack.
A new generic approach for scoping HTA. Iris Pasternack.A new generic approach for scoping HTA. Iris Pasternack.
A new generic approach for scoping HTA. Iris Pasternack.
HTAi Bilbao 2012
 
Machine learning in biology
Machine learning in biologyMachine learning in biology
Machine learning in biology
Pranavathiyani G
 
transformers_multimodal_ehr.pdf
transformers_multimodal_ehr.pdftransformers_multimodal_ehr.pdf
transformers_multimodal_ehr.pdf
Paris Women in Machine Learning and Data Science
 
Machine Learning in Biology and Why It Doesn't Make Sense - Theo Knijnenburg,...
Machine Learning in Biology and Why It Doesn't Make Sense - Theo Knijnenburg,...Machine Learning in Biology and Why It Doesn't Make Sense - Theo Knijnenburg,...
Machine Learning in Biology and Why It Doesn't Make Sense - Theo Knijnenburg,...
Seattle DAML meetup
 
International Journal of Biometrics and Bioinformatics(IJBB) Volume (4) Issue...
International Journal of Biometrics and Bioinformatics(IJBB) Volume (4) Issue...International Journal of Biometrics and Bioinformatics(IJBB) Volume (4) Issue...
International Journal of Biometrics and Bioinformatics(IJBB) Volume (4) Issue...CSCJournals
 
The Envisia Genomic Classifier
The Envisia Genomic ClassifierThe Envisia Genomic Classifier
The Envisia Genomic Classifier
Phil J. Morrison
 
Evaluation of the clinical value of biomarkers for risk prediction
Evaluation of the clinical value of biomarkers for risk predictionEvaluation of the clinical value of biomarkers for risk prediction
Evaluation of the clinical value of biomarkers for risk prediction
Ewout Steyerberg
 
Bio ontology drtc-seminar_anwesha
Bio ontology drtc-seminar_anweshaBio ontology drtc-seminar_anwesha
Bio ontology drtc-seminar_anwesha
anwesha bhattacharya
 
Biostatistics and its importance to Biologist
Biostatistics and its importance to BiologistBiostatistics and its importance to Biologist
Biostatistics and its importance to Biologist
Walid Nabil Saleh, DBA, MIBA, B.sc of Elec.Eng.
 
Predicting heart failure using a wrapper-based feature selection
Predicting heart failure using a wrapper-based feature selectionPredicting heart failure using a wrapper-based feature selection
Predicting heart failure using a wrapper-based feature selection
nooriasukmaningtyas
 
Bioengineered 3D Co culture Lung In Vitro Models: Platforms to Integrate Cell...
Bioengineered 3D Co culture Lung In Vitro Models: Platforms to Integrate Cell...Bioengineered 3D Co culture Lung In Vitro Models: Platforms to Integrate Cell...
Bioengineered 3D Co culture Lung In Vitro Models: Platforms to Integrate Cell...
Ken Rogan
 
Pathomics, Clinical Studies, and Cancer Surveillance
Pathomics, Clinical Studies, and Cancer SurveillancePathomics, Clinical Studies, and Cancer Surveillance
Pathomics, Clinical Studies, and Cancer Surveillance
Joel Saltz
 
14 00-20171207 rance-piv_c
14 00-20171207 rance-piv_c14 00-20171207 rance-piv_c
14 00-20171207 rance-piv_c
Bertrand Tavitian
 

Similar to End-to-end Fine-grained Neural Entity Recognition of Patients, Interventions, Outcomes (20)

DISTANT-CTO: A Zero Cost, Distantly Supervised Approach to Improve Low-Resour...
DISTANT-CTO: A Zero Cost, Distantly Supervised Approach to Improve Low-Resour...DISTANT-CTO: A Zero Cost, Distantly Supervised Approach to Improve Low-Resour...
DISTANT-CTO: A Zero Cost, Distantly Supervised Approach to Improve Low-Resour...
 
Conferencia del Dr. Ander Ramos en #Perspectives2015
Conferencia del Dr. Ander Ramos en #Perspectives2015Conferencia del Dr. Ander Ramos en #Perspectives2015
Conferencia del Dr. Ander Ramos en #Perspectives2015
 
EUSFLAT 2019: explainable neuro fuzzy recurrent neural network to predict col...
EUSFLAT 2019: explainable neuro fuzzy recurrent neural network to predict col...EUSFLAT 2019: explainable neuro fuzzy recurrent neural network to predict col...
EUSFLAT 2019: explainable neuro fuzzy recurrent neural network to predict col...
 
A review on early hospital mortality prediction using vital signals
A review on early hospital mortality prediction using vital signalsA review on early hospital mortality prediction using vital signals
A review on early hospital mortality prediction using vital signals
 
SNOMED CT concept model for molecular pathology_final.pptx
SNOMED CT concept model for molecular pathology_final.pptxSNOMED CT concept model for molecular pathology_final.pptx
SNOMED CT concept model for molecular pathology_final.pptx
 
Weakly supervised PICO information extraction using Snorkel
Weakly supervised PICO information extraction using SnorkelWeakly supervised PICO information extraction using Snorkel
Weakly supervised PICO information extraction using Snorkel
 
Real-time model for adaptive radiation therapy a biomechanical approach using...
Real-time model for adaptive radiation therapy a biomechanical approach using...Real-time model for adaptive radiation therapy a biomechanical approach using...
Real-time model for adaptive radiation therapy a biomechanical approach using...
 
A new generic approach for scoping HTA. Iris Pasternack.
A new generic approach for scoping HTA. Iris Pasternack.A new generic approach for scoping HTA. Iris Pasternack.
A new generic approach for scoping HTA. Iris Pasternack.
 
Machine learning in biology
Machine learning in biologyMachine learning in biology
Machine learning in biology
 
transformers_multimodal_ehr.pdf
transformers_multimodal_ehr.pdftransformers_multimodal_ehr.pdf
transformers_multimodal_ehr.pdf
 
Machine Learning in Biology and Why It Doesn't Make Sense - Theo Knijnenburg,...
Machine Learning in Biology and Why It Doesn't Make Sense - Theo Knijnenburg,...Machine Learning in Biology and Why It Doesn't Make Sense - Theo Knijnenburg,...
Machine Learning in Biology and Why It Doesn't Make Sense - Theo Knijnenburg,...
 
International Journal of Biometrics and Bioinformatics(IJBB) Volume (4) Issue...
International Journal of Biometrics and Bioinformatics(IJBB) Volume (4) Issue...International Journal of Biometrics and Bioinformatics(IJBB) Volume (4) Issue...
International Journal of Biometrics and Bioinformatics(IJBB) Volume (4) Issue...
 
The Envisia Genomic Classifier
The Envisia Genomic ClassifierThe Envisia Genomic Classifier
The Envisia Genomic Classifier
 
Evaluation of the clinical value of biomarkers for risk prediction
Evaluation of the clinical value of biomarkers for risk predictionEvaluation of the clinical value of biomarkers for risk prediction
Evaluation of the clinical value of biomarkers for risk prediction
 
Bio ontology drtc-seminar_anwesha
Bio ontology drtc-seminar_anweshaBio ontology drtc-seminar_anwesha
Bio ontology drtc-seminar_anwesha
 
Biostatistics and its importance to Biologist
Biostatistics and its importance to BiologistBiostatistics and its importance to Biologist
Biostatistics and its importance to Biologist
 
Predicting heart failure using a wrapper-based feature selection
Predicting heart failure using a wrapper-based feature selectionPredicting heart failure using a wrapper-based feature selection
Predicting heart failure using a wrapper-based feature selection
 
Bioengineered 3D Co culture Lung In Vitro Models: Platforms to Integrate Cell...
Bioengineered 3D Co culture Lung In Vitro Models: Platforms to Integrate Cell...Bioengineered 3D Co culture Lung In Vitro Models: Platforms to Integrate Cell...
Bioengineered 3D Co culture Lung In Vitro Models: Platforms to Integrate Cell...
 
Pathomics, Clinical Studies, and Cancer Surveillance
Pathomics, Clinical Studies, and Cancer SurveillancePathomics, Clinical Studies, and Cancer Surveillance
Pathomics, Clinical Studies, and Cancer Surveillance
 
14 00-20171207 rance-piv_c
14 00-20171207 rance-piv_c14 00-20171207 rance-piv_c
14 00-20171207 rance-piv_c
 

Recently uploaded

1.Seydhcuxhxyxhccuuxuxyxyxmisolids 2019.pptx
1.Seydhcuxhxyxhccuuxuxyxyxmisolids 2019.pptx1.Seydhcuxhxyxhccuuxuxyxyxmisolids 2019.pptx
1.Seydhcuxhxyxhccuuxuxyxyxmisolids 2019.pptx
Tiktokethiodaily
 
The affect of service quality and online reviews on customer loyalty in the E...
The affect of service quality and online reviews on customer loyalty in the E...The affect of service quality and online reviews on customer loyalty in the E...
The affect of service quality and online reviews on customer loyalty in the E...
jerlynmaetalle
 
一比一原版(CU毕业证)卡尔顿大学毕业证成绩单
一比一原版(CU毕业证)卡尔顿大学毕业证成绩单一比一原版(CU毕业证)卡尔顿大学毕业证成绩单
一比一原版(CU毕业证)卡尔顿大学毕业证成绩单
yhkoc
 
一比一原版(CBU毕业证)不列颠海角大学毕业证成绩单
一比一原版(CBU毕业证)不列颠海角大学毕业证成绩单一比一原版(CBU毕业证)不列颠海角大学毕业证成绩单
一比一原版(CBU毕业证)不列颠海角大学毕业证成绩单
nscud
 
Machine learning and optimization techniques for electrical drives.pptx
Machine learning and optimization techniques for electrical drives.pptxMachine learning and optimization techniques for electrical drives.pptx
Machine learning and optimization techniques for electrical drives.pptx
balafet
 
一比一原版(RUG毕业证)格罗宁根大学毕业证成绩单
一比一原版(RUG毕业证)格罗宁根大学毕业证成绩单一比一原版(RUG毕业证)格罗宁根大学毕业证成绩单
一比一原版(RUG毕业证)格罗宁根大学毕业证成绩单
vcaxypu
 
原版制作(Deakin毕业证书)迪肯大学毕业证学位证一模一样
原版制作(Deakin毕业证书)迪肯大学毕业证学位证一模一样原版制作(Deakin毕业证书)迪肯大学毕业证学位证一模一样
原版制作(Deakin毕业证书)迪肯大学毕业证学位证一模一样
u86oixdj
 
一比一原版(Bradford毕业证书)布拉德福德大学毕业证如何办理
一比一原版(Bradford毕业证书)布拉德福德大学毕业证如何办理一比一原版(Bradford毕业证书)布拉德福德大学毕业证如何办理
一比一原版(Bradford毕业证书)布拉德福德大学毕业证如何办理
mbawufebxi
 
一比一原版(UMich毕业证)密歇根大学|安娜堡分校毕业证成绩单
一比一原版(UMich毕业证)密歇根大学|安娜堡分校毕业证成绩单一比一原版(UMich毕业证)密歇根大学|安娜堡分校毕业证成绩单
一比一原版(UMich毕业证)密歇根大学|安娜堡分校毕业证成绩单
ewymefz
 
Data Centers - Striving Within A Narrow Range - Research Report - MCG - May 2...
Data Centers - Striving Within A Narrow Range - Research Report - MCG - May 2...Data Centers - Striving Within A Narrow Range - Research Report - MCG - May 2...
Data Centers - Striving Within A Narrow Range - Research Report - MCG - May 2...
pchutichetpong
 
Sample_Global Non-invasive Prenatal Testing (NIPT) Market, 2019-2030.pdf
Sample_Global Non-invasive Prenatal Testing (NIPT) Market, 2019-2030.pdfSample_Global Non-invasive Prenatal Testing (NIPT) Market, 2019-2030.pdf
Sample_Global Non-invasive Prenatal Testing (NIPT) Market, 2019-2030.pdf
Linda486226
 
FP Growth Algorithm and its Applications
FP Growth Algorithm and its ApplicationsFP Growth Algorithm and its Applications
FP Growth Algorithm and its Applications
MaleehaSheikh2
 
Criminal IP - Threat Hunting Webinar.pdf
Criminal IP - Threat Hunting Webinar.pdfCriminal IP - Threat Hunting Webinar.pdf
Criminal IP - Threat Hunting Webinar.pdf
Criminal IP
 
Ch03-Managing the Object-Oriented Information Systems Project a.pdf
Ch03-Managing the Object-Oriented Information Systems Project a.pdfCh03-Managing the Object-Oriented Information Systems Project a.pdf
Ch03-Managing the Object-Oriented Information Systems Project a.pdf
haila53
 
一比一原版(UIUC毕业证)伊利诺伊大学|厄巴纳-香槟分校毕业证如何办理
一比一原版(UIUC毕业证)伊利诺伊大学|厄巴纳-香槟分校毕业证如何办理一比一原版(UIUC毕业证)伊利诺伊大学|厄巴纳-香槟分校毕业证如何办理
一比一原版(UIUC毕业证)伊利诺伊大学|厄巴纳-香槟分校毕业证如何办理
ahzuo
 
一比一原版(TWU毕业证)西三一大学毕业证成绩单
一比一原版(TWU毕业证)西三一大学毕业证成绩单一比一原版(TWU毕业证)西三一大学毕业证成绩单
一比一原版(TWU毕业证)西三一大学毕业证成绩单
ocavb
 
一比一原版(QU毕业证)皇后大学毕业证成绩单
一比一原版(QU毕业证)皇后大学毕业证成绩单一比一原版(QU毕业证)皇后大学毕业证成绩单
一比一原版(QU毕业证)皇后大学毕业证成绩单
enxupq
 
一比一原版(NYU毕业证)纽约大学毕业证成绩单
一比一原版(NYU毕业证)纽约大学毕业证成绩单一比一原版(NYU毕业证)纽约大学毕业证成绩单
一比一原版(NYU毕业证)纽约大学毕业证成绩单
ewymefz
 
一比一原版(UofM毕业证)明尼苏达大学毕业证成绩单
一比一原版(UofM毕业证)明尼苏达大学毕业证成绩单一比一原版(UofM毕业证)明尼苏达大学毕业证成绩单
一比一原版(UofM毕业证)明尼苏达大学毕业证成绩单
ewymefz
 
一比一原版(Deakin毕业证书)迪肯大学毕业证如何办理
一比一原版(Deakin毕业证书)迪肯大学毕业证如何办理一比一原版(Deakin毕业证书)迪肯大学毕业证如何办理
一比一原版(Deakin毕业证书)迪肯大学毕业证如何办理
oz8q3jxlp
 

Recently uploaded (20)

1.Seydhcuxhxyxhccuuxuxyxyxmisolids 2019.pptx
1.Seydhcuxhxyxhccuuxuxyxyxmisolids 2019.pptx1.Seydhcuxhxyxhccuuxuxyxyxmisolids 2019.pptx
1.Seydhcuxhxyxhccuuxuxyxyxmisolids 2019.pptx
 
The affect of service quality and online reviews on customer loyalty in the E...
The affect of service quality and online reviews on customer loyalty in the E...The affect of service quality and online reviews on customer loyalty in the E...
The affect of service quality and online reviews on customer loyalty in the E...
 
一比一原版(CU毕业证)卡尔顿大学毕业证成绩单
一比一原版(CU毕业证)卡尔顿大学毕业证成绩单一比一原版(CU毕业证)卡尔顿大学毕业证成绩单
一比一原版(CU毕业证)卡尔顿大学毕业证成绩单
 
一比一原版(CBU毕业证)不列颠海角大学毕业证成绩单
一比一原版(CBU毕业证)不列颠海角大学毕业证成绩单一比一原版(CBU毕业证)不列颠海角大学毕业证成绩单
一比一原版(CBU毕业证)不列颠海角大学毕业证成绩单
 
Machine learning and optimization techniques for electrical drives.pptx
Machine learning and optimization techniques for electrical drives.pptxMachine learning and optimization techniques for electrical drives.pptx
Machine learning and optimization techniques for electrical drives.pptx
 
一比一原版(RUG毕业证)格罗宁根大学毕业证成绩单
一比一原版(RUG毕业证)格罗宁根大学毕业证成绩单一比一原版(RUG毕业证)格罗宁根大学毕业证成绩单
一比一原版(RUG毕业证)格罗宁根大学毕业证成绩单
 
原版制作(Deakin毕业证书)迪肯大学毕业证学位证一模一样
原版制作(Deakin毕业证书)迪肯大学毕业证学位证一模一样原版制作(Deakin毕业证书)迪肯大学毕业证学位证一模一样
原版制作(Deakin毕业证书)迪肯大学毕业证学位证一模一样
 
一比一原版(Bradford毕业证书)布拉德福德大学毕业证如何办理
一比一原版(Bradford毕业证书)布拉德福德大学毕业证如何办理一比一原版(Bradford毕业证书)布拉德福德大学毕业证如何办理
一比一原版(Bradford毕业证书)布拉德福德大学毕业证如何办理
 
一比一原版(UMich毕业证)密歇根大学|安娜堡分校毕业证成绩单
一比一原版(UMich毕业证)密歇根大学|安娜堡分校毕业证成绩单一比一原版(UMich毕业证)密歇根大学|安娜堡分校毕业证成绩单
一比一原版(UMich毕业证)密歇根大学|安娜堡分校毕业证成绩单
 
Data Centers - Striving Within A Narrow Range - Research Report - MCG - May 2...
Data Centers - Striving Within A Narrow Range - Research Report - MCG - May 2...Data Centers - Striving Within A Narrow Range - Research Report - MCG - May 2...
Data Centers - Striving Within A Narrow Range - Research Report - MCG - May 2...
 
Sample_Global Non-invasive Prenatal Testing (NIPT) Market, 2019-2030.pdf
Sample_Global Non-invasive Prenatal Testing (NIPT) Market, 2019-2030.pdfSample_Global Non-invasive Prenatal Testing (NIPT) Market, 2019-2030.pdf
Sample_Global Non-invasive Prenatal Testing (NIPT) Market, 2019-2030.pdf
 
FP Growth Algorithm and its Applications
FP Growth Algorithm and its ApplicationsFP Growth Algorithm and its Applications
FP Growth Algorithm and its Applications
 
Criminal IP - Threat Hunting Webinar.pdf
Criminal IP - Threat Hunting Webinar.pdfCriminal IP - Threat Hunting Webinar.pdf
Criminal IP - Threat Hunting Webinar.pdf
 
Ch03-Managing the Object-Oriented Information Systems Project a.pdf
Ch03-Managing the Object-Oriented Information Systems Project a.pdfCh03-Managing the Object-Oriented Information Systems Project a.pdf
Ch03-Managing the Object-Oriented Information Systems Project a.pdf
 
一比一原版(UIUC毕业证)伊利诺伊大学|厄巴纳-香槟分校毕业证如何办理
一比一原版(UIUC毕业证)伊利诺伊大学|厄巴纳-香槟分校毕业证如何办理一比一原版(UIUC毕业证)伊利诺伊大学|厄巴纳-香槟分校毕业证如何办理
一比一原版(UIUC毕业证)伊利诺伊大学|厄巴纳-香槟分校毕业证如何办理
 
一比一原版(TWU毕业证)西三一大学毕业证成绩单
一比一原版(TWU毕业证)西三一大学毕业证成绩单一比一原版(TWU毕业证)西三一大学毕业证成绩单
一比一原版(TWU毕业证)西三一大学毕业证成绩单
 
一比一原版(QU毕业证)皇后大学毕业证成绩单
一比一原版(QU毕业证)皇后大学毕业证成绩单一比一原版(QU毕业证)皇后大学毕业证成绩单
一比一原版(QU毕业证)皇后大学毕业证成绩单
 
一比一原版(NYU毕业证)纽约大学毕业证成绩单
一比一原版(NYU毕业证)纽约大学毕业证成绩单一比一原版(NYU毕业证)纽约大学毕业证成绩单
一比一原版(NYU毕业证)纽约大学毕业证成绩单
 
一比一原版(UofM毕业证)明尼苏达大学毕业证成绩单
一比一原版(UofM毕业证)明尼苏达大学毕业证成绩单一比一原版(UofM毕业证)明尼苏达大学毕业证成绩单
一比一原版(UofM毕业证)明尼苏达大学毕业证成绩单
 
一比一原版(Deakin毕业证书)迪肯大学毕业证如何办理
一比一原版(Deakin毕业证书)迪肯大学毕业证如何办理一比一原版(Deakin毕业证书)迪肯大学毕业证如何办理
一比一原版(Deakin毕业证书)迪肯大学毕业证如何办理
 

End-to-end Fine-grained Neural Entity Recognition of Patients, Interventions, Outcomes

  • 1. Presenter: Anjani K. Dhrangadhariya Joint work with: G. Aguilar, T. Solorio, R. Hilfiker, H. Müller CLEF 2021, Bucharest, Romania, 23.09.2021 1 End-to-end Fine-grained Neural Entity Recognition of Patients, Interventions, Outcomes
  • 2. Introduction 2 …A semi-structured interview was used to obtain qualitative information on the effect of the aerobics intervention vs. conventional exercise. The convenience sample included 15 adult Oncology outpatients, 13 female and 2 male, ranging in age from 20 to 87. Quality of life, fatigue, and side-effects were the measured outcomes… Participant Intervention Comparator Outcome
  • 3. Introduction 3 …A semi-structured interview was used to obtain qualitative information on the effect of the aerobics intervention vs. conventional exercise. The convenience sample included 15 adult Oncology outpatients, 13 female and 2 male, ranging in age from 20 to 87. Quality of life, fatigue, and side-effects were the measured outcomes… P I C O aerobics intervention Quality of life fatigue conventional exercise 15 Adult, age from 20 to 87 Oncology outpatients 13 female and 2 male side-effects Sample size Age Condition Gender Physical Control Physical Physical Adverse effects Coarse-grained labels Fine-grained labels
  • 4. 4 P I C O aerobics intervention Quality of life fatigue conventional exercise 15 Adult, age from 20 to 87 Oncology outpatients 13 female and 2 male side-effects Sample size Age Condition Gender Physical Control Physical Physical Adverse effects Systematic Reviews (SR) Inclusion into the SR Exclude from the SR • Evidence-based clinical practice • Health policy-making Introduction
  • 5. Challenges • Coarse-grained PICO - Disagreement between human annotators • Fine-grained PICO subtypes - are neither clearly identified nor standardized as semantic units. • Domain-specific challenges • Naming conventions are not established in the health vs. pharma literature 5 Brockmeier, Austin J., et al. "Improving reference prioritisation with PICO recognition." BMC medical informatics and decision making 19.1 (2019): 1-14.
  • 6. Challenges • Sentence-level PICO extraction • Only available “large” entity annotation PICO dataset – EBM-PICO • ~5000 PICO annotated RCT abstract • Manually annotated – PICO and subtypes • Training and Test sets • Annotators: Medical students and physicians • Only available with multi-level annotation 6 Nye, Benjamin, et al. "A corpus with multi-level annotations of patients, interventions and outcomes to support language processing for medical literature." Proceedings of the conference. Association for Computational Linguistics. Meeting. Vol. 2018. NIH Public Access, 2018.
  • 7. State of the art Nye et al. Hand-engineered features Poor performance on the fine-grained PICO Ananiadou et al. Do not focus on the fine-grained recognition Beltagy et al. SciBERT domain-adaptation improved coarse-grained recognition. Zhang et al. Disease NER - Participant condition entity only Xu et al. Focus on patient demographics (gender, sex, ethnicity) alone Chung et al. Focus on the Intervention arms in the RCTs 7
  • 8. 1. Most studies concentrate on the coarse-grained PICO recognition but not on the extraction of PICO subtypes and are not end-to-end. • PICO recognition = Semi-automation • PICO subtype recognition = full automation 2. No focus on domain differences 3. No exploration of using coarse and fine-grained labels annotation together – Multi-task learning (MTL)? 8 Motivation
  • 9. Methodology: MTL Single-task learning Multi-task learning • Task A, Task B • Labeled dataset d(xa ,ya), d(xb ,yb) • Loss function La, Lb • ModelA, ModelB • Task A, Task B • Labeled dataset d(x,ya), d(x,yb) • A loss function L • Model Models trained separately and do not share the L and the internal parameters Models trained in unison and share the L and the internal parameters 9 • Train one neural network to do multiple related tasks where each task hopefully helps improve the performance on the other task. https://ruder.io/multi-task/index.html
  • 10. Methodology: MTL • The key idea is that one task (auxiliary task) can help another task (main task). • Assumption: Fine-grained entities nested under the coarse-grained entities = related tasks. • Mutual sources of inductive bias 10 Coarse-grained Participant labels Fine-grained Participant labels
  • 11. Methodology: Datasets 12 • EBM-PICO training set • 5000 PICO annotated RCT abstracts • PICO coarse- and fine-grained • Training set: 4,993 documents • Annotated by medical students EBM-PICO evaluation corpus Hilfiker corpus • 191 PICO documents (fine and coarse) • Annotated by physicians • Pharma domain • 153 PICO annotated (fine and coarse) • Annotated by the first author • Physiotherapy and Rehab domain https://www.ncbi.nlm.nih.gov/pmc/articles/PMC6174533/bin/NIHMS988059-supplement-Appendix.pdf Documents = RCT abstracts
  • 12. 13 Prepares the input text tokens by subword tokenizing them using BERT tokenizer Methodology: MTL architecture
  • 13. 14 Methodology: MTL architecture Used to extract deep contextualized BERT vectors from the tokenized text
  • 14. 15 Methodology: MTL architecture Average over the last four BERT layers are calculated and the representation is fed to the BiLSTM layer
  • 15. 16 Methodology: MTL architecture BiLSTM outputs are concatenated and self- attention is applied
  • 16. 17 Methodology: MTL architecture Emission probabilities from the self-attention layer are fed into the two separate CRF heads along with the true labels
  • 17. 18 Methodology: MTL architecture = Decoder layers separately learn the coarse- and fine- grained labels on the data and calculate individual losses 1. MTL 2. STL
  • 18. Experiments conducted 19 Experiment Model setup I BERT Linear II BERT LSTM CRF III BERT BiLSTM CRF IV BERT LSTM Atten CRF V BERT BiLSTM Atten CRF V BERT BiLSTM Multihead Atten CRF VI BERT BiLSTM Multihead Atten CRF (fine-)/Linear(coarse-)
  • 19. Experiments conducted 20 Experiment Model setup I BERT Linear II BERT LSTM CRF III BERT BiLSTM CRF IV BERT LSTM Atten CRF V BERT BiLSTM Atten CRF Ablation 1 BERT BiLSTM attn (on coarse) CRF Ablation 2 BERT BiLSTM attn (on fine) CRF V BERT BiLSTM Multihead Atten CRF VI BERT BiLSTM Multihead Atten CRF (fine-)/Linear(coarse-) 1. MTL 2. STL
  • 20. Results 21 Underline = Significant improvement compared to the counterpart MTL performs better (BOLD) for some experimental setups but it is not a significant improvement (UNDERLINE)
  • 22. Error analysis (EBM-PICO eval set) 23 • Intervention entity : Poor performance • “Psychological” and “Other” classes • Fewer annotations • Consistently confused between “Educational” and “Psychological” – Annotation exercise • Plenty PICO entities get classified as out-of-the- span label and vice versa. • Class overlap + Class imbalance
  • 23. Conclusion 1. Propose two end-to-end neural model setups for fine-grained PICO recognition. 2. Both model setups outperform the previous approaches for the fine-grained ``Participant'' and ``Outcome'' entity recognition. 3. MTL is a good alternative to STL but not directly. 4. We contribute a PICO annotated corpus for the physiotherapy domain and explore the domain difference • Our error analysis warrants rethinking of semantically solid class definitions for fine-grained PICO entities along with ontology development for the healthcare domain. 24
  • 24. Training setup 1. Epochs 15 2. Max. seq. length BERT 512 3. BERTs’ Last four layers summed 4. BERT not freezed 5. LSTM/BiLSTM hidden size – 512/1025 6. Optimizer – AdamW, Learning rate 5e-5 7. Gradients clipped to one. 8. Weighted cross-entropy loss using Sklearn suite 9. A single Tesla K80 GPU, PyTorch 25
  • 25. References • Boudin, Florian, et al. "Combining classifiers for robust PICO element detection." BMC medical informatics and decision making 10.1 (2010): 1-6. • Huang, Ke-Chun, et al. "Classification of PICO elements by text features systematically extracted from PubMed abstracts." 2011 IEEE International Conference on Granular Computing. IEEE, 2011. • Jin, Di, and Peter Szolovits. "Pico element detection in medical text via long short-term memory neural networks." Proceedings of the BioNLP 2018 workshop. 2018. • Beltagy, Iz, Kyle Lo, and Arman Cohan. "Scibert: A pretrained language model for scientific text." arXiv preprint arXiv:1903.10676 (2019). • Zhang, Tengteng, et al. "Unlocking the power of deep pico extraction: Step-wise medical ner identification." arXiv preprint arXiv:2005.06601 (2020). • Xu, Rong, et al. "Extracting subject demographic information from abstracts of randomized clinical trial reports." Medinfo 2007: Proceedings of the 12th World Congress on Health (Medical) Informatics; Building Sustainable Health Systems. IOS Press, 2007. 26
  • 26. Thank you for your attention Questions? Anjani Dhrangadhariya anjani.dhrangadhariya@hevs.ch https://www.linkedin.com/in/anjani-dhrangadhariya/ More information http://medgift.hevs.ch/wordpress/ https://www.examode.eu/ 27