Deep-learning based detection of eosinophilic esophagitis.

Researchers

Journal

Modalities

Models

Abstract

For eosinophilic esophagitis (EoE) a substantial diagnostic delay is still a clinically relevant phenomenon. Deep learning-based algorithms have demonstrated potential in medical image analysis. Here we establish a convolutional neuronal network (CNN)-based approach that can distinguish EoE from normal findings and candida esophagitis.
We trained and tested a CNN using 484 real-world endoscopic images from 134 subjects consisting of three classes (normal, EoE, and candidiasis). Images were split into two completely independent datasets. The proposed approach was evaluated against three trainee endoscopists on the test set. Model-explainability was enhanced by deep Taylor decomposition.
Global accuracy (0.915 [0.880-0.940]), sensitivity (0.871 [0.819-0.910]) and specificity (0.936 [0.910-0.955]) were significantly higher than for endoscopists on the test set. Global area under the ROC curve was 0.966 [0.954-0.975]. Results were highly reproducible. Explainability analysis found that the algorithm identified characteristic signs also used by endoscopists.
Complex endoscopic classification tasks including more than two classes can be solved by CNN-based algorithms. Thus, our algorithm (https://ccb-test.cs.uni-saarland.de/EoE/) may assist clinicians in making the diagnosis of EoE.
Thieme. All rights reserved.