Robust and Interpretable Convolutional Neural Networks to Detect Glaucoma in Optical Coherence Tomography Images
Model .h5 files and .pb files for robustly detecting glaucoma from optical coherence tomography (OCT) images and for interpretability analysis via testing with concept activation vectors (TCAVs by Been Kim et al.). Further details described in paper "Robust and Interpretable Convolutional Neural Networks to Detect Glaucoma in OCT Images" in preparation/under review.
Included are four .h5 files corresponding to the end-to-end deep learning OCT-fine-tuned Convolutional Neural Networks described in our paper. These models can be loaded using Keras and applied to detect glaucoma in OCT images (retinal nerve fiber layer probability maps). Also included are the .h5, .pb, and .txt files corresponding to the InceptionV3 + FC model and OCT-concept labels used for interpretability analysis via TCAVs as described in our paper. This model can be used via the wrapper class implemented on our forked TCAV repository included in the ‘Run TCAV’ jupyter notebooks located in our GitHub repository: http://github.com/LIINC/TCAV4OCT under src/TCAVRandomConcepts10 and under src/TCAVRandomConcepts160.
- End-to-End Deep Learning Models, TCAV Model, OCT-Concept Labels e2eModels_TCAVModel_Labels.zip (4.69 GB)
Open Access dataset files are accessible to all logged in users. Don't have a login? Create a free IEEE account. IEEE Membership is not required.