Deep learning holds great promise for the medical field. A computational model can assist doctors with references or prioritization as they triage or diagnose patients. When professional expertise is scarce, the model can make a best guess based on the statistics of all the patient history it has seen. In a concrete example from 2017, Pranav Rajpurkar et al from the Stanford ML Group developed CheXNet, a deep convolutional neural network which—in certain controlled contexts—claims to diagnose pneumonia from chest x-rays more accurately than an average radiologist.
In this report, I explore the open dataset of chest x-rays, train simple baselines on a 5% subsample, and outline the next steps to model improvement in a real-world medical scenario: one where data is scarce, noisy, and long-tailed and where the model must to be maximally explainable.