Keywords: Model interpretability, Medical image analysis
Abstract: Despite the progress of deep learning on medical imaging, there is still not a true understanding of what networks learn and of how decisions are reached. Here, we address this by proposing a Visualized Neural-backed Decision Tree for Medical image analysis, Me-NDT. It is a CNN with a tree-based structure template that allows for both classification and visualization of firing neurons, thus offering interpretability. We also introduce node and path losses that allow Me-NDT to consider the entire path instead of isolated nodes. Our experiments on brain CT and chest radiographs outperform all baselines. Overall, Me-NDT is a lighter, comprehensively explanatory model, of great value for clinical practice.
Paper Type: both
Primary Subject Area: Interpretability and Explainable AI
Secondary Subject Area: Detection and Diagnosis
Paper Status: original work, not submitted yet
Source Code Url: Our code needs to be sorted out before it can be made public.
Data Set Url: Brain CT (provided by RSNA challenge): https://www.kaggle.com/c/rsna-intracranial-hemorrhage-detection/overview
Registration: I acknowledge that publication of this at MIDL and in the proceedings requires at least one of the authors to register and present the work during the conference.
Authorship: I confirm that I am the author of this work and that it has not been submitted to another publication before.