Skip to content

Latest commit

 

History

History
29 lines (20 loc) · 2.15 KB

README.md

File metadata and controls

29 lines (20 loc) · 2.15 KB
Raw

TCAV4OCT

Robust and Interpretable CNNs for Glaucoma Detection from OCT Images

Code and Scripts for robust end-to-end deep learning models and interpretabilty analysis using Testing with Concept Activation Vectors (TCAVs, Been Kim, et al.) and corroboration with expert eye tracking, as described in paper: "Robust and Interpretable Convolutional Neural Networks to Detect Glaucoma in Optical Coherence Tomography Images" published in IEEE Transactions on Biomedical Engineering: https://ieeexplore.ieee.org/document/9286420.

by Kaveri A. Thakoor, Sharath Koorathota, Donald C. Hood, and Paul Sajda

src:

  1. end2endDLModels: contains jupyter notebooks/python code for robust end-to-end deep learning models fine-tuned on OCT data
  2. eyeTracking: contains script for generating expert eye fixation heatmaps superimposed on OCT reports and for computing fixation density per Area of Interest, contains modified plotting toolkit (PyGazeAnalyzer)
  3. TCAVRandomConcepts10: scripts for TCAV interpretability analysis using 10 random concepts
  4. TCAVRandomConcepts160: scripts for TCAV interpretability analysis using 160 random concepts

models: Info for accessing saved models available on IEEE DataPort

results: Results file for 160 random concepts TCAV experiment

doc: Info on 150 ImageNet concepts used for TCAV experiment, additional OCT concepts and hand-selected random concepts listed in scripts under src/TCAVRandomConcepts10

References (and more in paper described above):

  1. F. Chollet, Deep Learning with Python. Manning Publications, 2018.
  2. B. Kim, M. Wattenberg, J. Gilmer, C. Cai, J. Wexler, F. Viegas, and R. Sayres. "Interpretability beyond feature attribution: Quantitative testing with concept activation vectors (TCAV)", International Conference on Machine Learning, 2017.
  3. E.S. Dalmaijer, S. Mathôt, and S. Van der Stigchel, "PyGaze: An open-source, cross-platform toolbox for minimal-effort programming of eyetracking experiments." Behavior research methods, 46(4), pp.913-921, 2014.
  4. J. Deng, W. Dong, R. Socher, L. Li, K. Li, F. Li. "Imagenet: A large-scale hierarchical image database." IEEE Conference on Computer Vision and Pattern Recognition, pp. 248-255, 2009.