Topological Machine Learning for Low Data Medical Imaging

Brighton Nuwagira, Caner Korkmaz, Philmore Koung, Baris Coskunuzer
Proceedings of the 4th Machine Learning for Health Symposium, PMLR 259:824-838, 2025.

Abstract

Deep Learning (DL) has revolutionized medical image analysis by providing automated techniques to extract valuable insights from large datasets. However, challenges such as interpretability and reliance on extensive labeled data persist. Topological Data Analysis (TDA) has emerged as a complementary approach that captures underlying topological structures in data, potentially enhancing the performance of DL models. In this paper, we present a comprehensive evaluation of TDA methods for computer-aided diagnosis from two perspectives. First, we examine the effectiveness of topological methods in data-limited settings by comparing the standalone performance of DL models, TDA approaches, and their fusion. Our results demonstrate that integrating topological features into DL models significantly improves performance when labeled data are scarce. Second, we assess the standalone performance of TDA methods in data-rich environments using the MedMNIST collection, which includes over 600K images across 12 2D and 6 3D medical imaging datasets. Our experiments reveal that while TDA methods do not outperform DL models on 2D datasets, they achieve competitive results on 3D imaging tasks. These findings suggest that the fusion of TDA and DL methods can enhance the accuracy and robustness of computer-aided diagnosis, particularly in low-data or 3D imaging scenarios.

Cite this Paper


BibTeX
@InProceedings{pmlr-v259-nuwagira25a, title = {Topological Machine Learning for Low Data Medical Imaging}, author = {Nuwagira, Brighton and Korkmaz, Caner and Koung, Philmore and Coskunuzer, Baris}, booktitle = {Proceedings of the 4th Machine Learning for Health Symposium}, pages = {824--838}, year = {2025}, editor = {Hegselmann, Stefan and Zhou, Helen and Healey, Elizabeth and Chang, Trenton and Ellington, Caleb and Mhasawade, Vishwali and Tonekaboni, Sana and Argaw, Peniel and Zhang, Haoran}, volume = {259}, series = {Proceedings of Machine Learning Research}, month = {15--16 Dec}, publisher = {PMLR}, pdf = {https://raw.githubusercontent.com/mlresearch/v259/main/assets/nuwagira25a/nuwagira25a.pdf}, url = {https://proceedings.mlr.press/v259/nuwagira25a.html}, abstract = {Deep Learning (DL) has revolutionized medical image analysis by providing automated techniques to extract valuable insights from large datasets. However, challenges such as interpretability and reliance on extensive labeled data persist. Topological Data Analysis (TDA) has emerged as a complementary approach that captures underlying topological structures in data, potentially enhancing the performance of DL models. In this paper, we present a comprehensive evaluation of TDA methods for computer-aided diagnosis from two perspectives. First, we examine the effectiveness of topological methods in data-limited settings by comparing the standalone performance of DL models, TDA approaches, and their fusion. Our results demonstrate that integrating topological features into DL models significantly improves performance when labeled data are scarce. Second, we assess the standalone performance of TDA methods in data-rich environments using the MedMNIST collection, which includes over 600K images across 12 2D and 6 3D medical imaging datasets. Our experiments reveal that while TDA methods do not outperform DL models on 2D datasets, they achieve competitive results on 3D imaging tasks. These findings suggest that the fusion of TDA and DL methods can enhance the accuracy and robustness of computer-aided diagnosis, particularly in low-data or 3D imaging scenarios.} }
Endnote
%0 Conference Paper %T Topological Machine Learning for Low Data Medical Imaging %A Brighton Nuwagira %A Caner Korkmaz %A Philmore Koung %A Baris Coskunuzer %B Proceedings of the 4th Machine Learning for Health Symposium %C Proceedings of Machine Learning Research %D 2025 %E Stefan Hegselmann %E Helen Zhou %E Elizabeth Healey %E Trenton Chang %E Caleb Ellington %E Vishwali Mhasawade %E Sana Tonekaboni %E Peniel Argaw %E Haoran Zhang %F pmlr-v259-nuwagira25a %I PMLR %P 824--838 %U https://proceedings.mlr.press/v259/nuwagira25a.html %V 259 %X Deep Learning (DL) has revolutionized medical image analysis by providing automated techniques to extract valuable insights from large datasets. However, challenges such as interpretability and reliance on extensive labeled data persist. Topological Data Analysis (TDA) has emerged as a complementary approach that captures underlying topological structures in data, potentially enhancing the performance of DL models. In this paper, we present a comprehensive evaluation of TDA methods for computer-aided diagnosis from two perspectives. First, we examine the effectiveness of topological methods in data-limited settings by comparing the standalone performance of DL models, TDA approaches, and their fusion. Our results demonstrate that integrating topological features into DL models significantly improves performance when labeled data are scarce. Second, we assess the standalone performance of TDA methods in data-rich environments using the MedMNIST collection, which includes over 600K images across 12 2D and 6 3D medical imaging datasets. Our experiments reveal that while TDA methods do not outperform DL models on 2D datasets, they achieve competitive results on 3D imaging tasks. These findings suggest that the fusion of TDA and DL methods can enhance the accuracy and robustness of computer-aided diagnosis, particularly in low-data or 3D imaging scenarios.
APA
Nuwagira, B., Korkmaz, C., Koung, P. & Coskunuzer, B.. (2025). Topological Machine Learning for Low Data Medical Imaging. Proceedings of the 4th Machine Learning for Health Symposium, in Proceedings of Machine Learning Research 259:824-838 Available from https://proceedings.mlr.press/v259/nuwagira25a.html.

Related Material