Interpretable histopathology-based prediction of disease relevant features in Inflammatory Bowel Disease biopsies using weakly-supervised deep learning

Ricardo Mokhtari, Azam Hamidinekoo, Daniel James Sutton, Arthur Lewis, Bastian Angermann, Ulf Gehrmann, Pål Lundin, Hibret Adissu, Junmei Cairns, Jessica Neisen, Emon Khan, Daniel Marks, Nia Khachapuridze, Talha Qaiser, Nikolay Burlutskiy
Medical Imaging with Deep Learning, PMLR 227:479-495, 2024.

Abstract

Crohn’s Disease (CD) and Ulcerative Colitis (UC) are the two main Inflammatory Bowel Disease (IBD) types. We developed interpretable deep learning models to identify histolog- ical disease features for both CD and UC using only endoscopic labels. We explored fine- tuning and end-to-end training of two state-of-the-art self-supervised models for predicting three different endoscopic categories (i) CD vs UC (AUC=0.87), (ii) normal vs lesional (AUC=0.81), (iii) low vs high disease severity score (AUC=0.80). With the support of a pathologist, we explored the relationship between endoscopic labels, model predictions and histological evaluations qualitatively and quantitatively and identified cases where the pathologist’s descriptions of inflammation were consistent with regions of high attention. In parallel, we used a model trained on the Colon Nuclei Identification and Counting (CoNIC) dataset to predict and explore 6 cell populations. We observed consistency between areas enriched with the predicted immune cells in biopsies and the pathologist’s feedback on the attention maps. Finally, we identified several cell level features indicative of disease severity in CD and UC. These models can enhance our understanding about the pathology behind IBD and can shape our strategies for patient stratification in clinical trials.

Cite this Paper

BibTeX
@InProceedings{pmlr-v227-mokhtari24a, title = {Interpretable histopathology-based prediction of disease relevant features in Inflammatory Bowel Disease biopsies using weakly-supervised deep learning}, author = {Mokhtari, Ricardo and Hamidinekoo, Azam and Sutton, Daniel James and Lewis, Arthur and Angermann, Bastian and Gehrmann, Ulf and Lundin, P\r{a}l and Adissu, Hibret and Cairns, Junmei and Neisen, Jessica and Khan, Emon and Marks, Daniel and Khachapuridze, Nia and Qaiser, Talha and Burlutskiy, Nikolay}, booktitle = {Medical Imaging with Deep Learning}, pages = {479--495}, year = {2024}, editor = {Oguz, Ipek and Noble, Jack and Li, Xiaoxiao and Styner, Martin and Baumgartner, Christian and Rusu, Mirabela and Heinmann, Tobias and Kontos, Despina and Landman, Bennett and Dawant, Benoit}, volume = {227}, series = {Proceedings of Machine Learning Research}, month = {10--12 Jul}, publisher = {PMLR}, pdf = {https://proceedings.mlr.press/v227/mokhtari24a/mokhtari24a.pdf}, url = {https://proceedings.mlr.press/v227/mokhtari24a.html}, abstract = {Crohn’s Disease (CD) and Ulcerative Colitis (UC) are the two main Inflammatory Bowel Disease (IBD) types. We developed interpretable deep learning models to identify histolog- ical disease features for both CD and UC using only endoscopic labels. We explored fine- tuning and end-to-end training of two state-of-the-art self-supervised models for predicting three different endoscopic categories (i) CD vs UC (AUC=0.87), (ii) normal vs lesional (AUC=0.81), (iii) low vs high disease severity score (AUC=0.80). With the support of a pathologist, we explored the relationship between endoscopic labels, model predictions and histological evaluations qualitatively and quantitatively and identified cases where the pathologist’s descriptions of inflammation were consistent with regions of high attention. In parallel, we used a model trained on the Colon Nuclei Identification and Counting (CoNIC) dataset to predict and explore 6 cell populations. We observed consistency between areas enriched with the predicted immune cells in biopsies and the pathologist’s feedback on the attention maps. Finally, we identified several cell level features indicative of disease severity in CD and UC. These models can enhance our understanding about the pathology behind IBD and can shape our strategies for patient stratification in clinical trials.} }
Endnote
%0 Conference Paper %T Interpretable histopathology-based prediction of disease relevant features in Inflammatory Bowel Disease biopsies using weakly-supervised deep learning %A Ricardo Mokhtari %A Azam Hamidinekoo %A Daniel James Sutton %A Arthur Lewis %A Bastian Angermann %A Ulf Gehrmann %A Pål Lundin %A Hibret Adissu %A Junmei Cairns %A Jessica Neisen %A Emon Khan %A Daniel Marks %A Nia Khachapuridze %A Talha Qaiser %A Nikolay Burlutskiy %B Medical Imaging with Deep Learning %C Proceedings of Machine Learning Research %D 2024 %E Ipek Oguz %E Jack Noble %E Xiaoxiao Li %E Martin Styner %E Christian Baumgartner %E Mirabela Rusu %E Tobias Heinmann %E Despina Kontos %E Bennett Landman %E Benoit Dawant %F pmlr-v227-mokhtari24a %I PMLR %P 479--495 %U https://proceedings.mlr.press/v227/mokhtari24a.html %V 227 %X Crohn’s Disease (CD) and Ulcerative Colitis (UC) are the two main Inflammatory Bowel Disease (IBD) types. We developed interpretable deep learning models to identify histolog- ical disease features for both CD and UC using only endoscopic labels. We explored fine- tuning and end-to-end training of two state-of-the-art self-supervised models for predicting three different endoscopic categories (i) CD vs UC (AUC=0.87), (ii) normal vs lesional (AUC=0.81), (iii) low vs high disease severity score (AUC=0.80). With the support of a pathologist, we explored the relationship between endoscopic labels, model predictions and histological evaluations qualitatively and quantitatively and identified cases where the pathologist’s descriptions of inflammation were consistent with regions of high attention. In parallel, we used a model trained on the Colon Nuclei Identification and Counting (CoNIC) dataset to predict and explore 6 cell populations. We observed consistency between areas enriched with the predicted immune cells in biopsies and the pathologist’s feedback on the attention maps. Finally, we identified several cell level features indicative of disease severity in CD and UC. These models can enhance our understanding about the pathology behind IBD and can shape our strategies for patient stratification in clinical trials.
APA
Mokhtari, R., Hamidinekoo, A., Sutton, D.J., Lewis, A., Angermann, B., Gehrmann, U., Lundin, P., Adissu, H., Cairns, J., Neisen, J., Khan, E., Marks, D., Khachapuridze, N., Qaiser, T. & Burlutskiy, N.. (2024). Interpretable histopathology-based prediction of disease relevant features in Inflammatory Bowel Disease biopsies using weakly-supervised deep learning. Medical Imaging with Deep Learning, in Proceedings of Machine Learning Research 227:479-495 Available from https://proceedings.mlr.press/v227/mokhtari24a.html.

Related Material