Interpretable Automated Diagnosis of Retinal Disease Using Deep OCT Analysis

  • 2022-01-13 18:56:50
  • Evan Wen, Max Ehrlich
  • 0


30 million Optical Coherence Tomography (OCT) imaging tests are issued everyyear to diagnose various retinal diseases, but accurate diagnosis of OCT scansrequires trained ophthalmologists who are still prone to making errors. Withbetter systems for diagnosis, many cases of vision loss caused by retinaldisease could be entirely avoided. In this work, we develop a novel deeplearning architecture for explainable, accurate classification of retinaldisease which achieves state-of-the-art accuracy. Furthermore, we place anemphasis on producing both qualitative and quantitative explanations of themodel's decisions. Our algorithm produces heatmaps indicating the exact regionsin the OCT scan the model focused on when making its decision. In combinationwith an OCT segmentation model, this allows us to produce quantitativebreakdowns of the specific retinal layers the model focused on for later reviewby an expert. Our work is the first to produce detailed quantitativeexplanations of the model's decisions in this way. Our combination of accuracyand interpretability can be clinically applied for better patient care.


Quick Read (beta)

loading the full paper ...