Robustness

Robust Glaucoma Detection at Multiple Locations

We developed end-to-end deep learning models which enable robust glaucoma detection from data collected at multiple locations by harnessing the power of natural-image pre-trained neural networks followed by fine-tuning on optical coherence tomography images. We are now in the process of evaluating these models for their value-added in the clinical workflow by measuring the impact on time and accuracy of clinical diagnosis when AI predictions are presented to clinicians alongside typical OCT data.

Read more here:

Thakoor, K., Koorathota, S., Hood, D., Sajda, P. “Robust and Interpretable Convolutional Neural Networks to Detect Glaucoma in Optical Coherence Tomography Images.” IEEE Transactions on Biomedical Engineering, 68(8), pp. 2456‑2466, August 2021. Early Access: https://ieeexplore.ieee.org/document/9286420, 8 December 2020.

Robust AMD Detection from Multiple Modalities

We achieved state-of-the-art 3-class Age-Related Macular Degeneration (AMD) detection from multiple imaging modalities (OCT, OCTA, high-definition 5-line 2D b-scans, and low-resolution 2D b-scans). We also achieved interpretability via Grad-CAMs and via comparing odds ratios of features of importance both for AI and for human experts. Variation in feature rank will help guide us as to how to improve AI and may help elucidate novel clinical features for accurate AMD detection.

Read more here:

Thakoor, K.A., Yao, J., Bordbar, D., Moussa, O., Lin, W., Sajda, P., Chen, R. “A Multimodal Deep Learning System to Distinguish Late Stages of AMD and to Compare Expert vs. AI Ocular Biomarkers.” Scientific Reports, 12(1), p.1‑11, 2022.