Artificial intelligence (AI) researchers and radiologists have recently reported AI systems that accurately detect COVID-19 in chest radiographs. However, the robustness of these systems remains unclear. Using state-of-the-art techniques in explainable AI, we demonstrate that recent deep learning systems to detect COVID-19 from chest radiographs rely on confounding factors rather than medical pathology, creating an alarming situation in which the systems appear accurate, but fail when tested in new hospitals.
While deep learning has shown promise in the domain of disease classification from medical images, models based on state-of-the-art convolutional neural network architectures often exhibit performance loss due to dataset shift. Models trained using data from one hospital system achieve high predictive performance when tested on data from the same hospital, but perform significantly worse when they are tested in different hospital systems. Furthermore, even within a given hospital system, deep learning models have been shown to depend on hospital- and patient-level confounders rather than meaningful pathology to make classifications. In order for these models to be safely deployed, we would like to ensure that they do not use confounding variables to make their classification, and that they will work well even when tested on images from hospitals that were not included in the training data. We attempt to address this problem in the context of pneumonia classification from chest radiographs. We propose an approach based on adversarial optimization, which allows us to learn more robust models that do not depend on confounders. Specifically, we demonstrate improved out-of-hospital generalization performance of a pneumonia classifier by training a model that is invariant to the view position of chest radiographs (anterior-posterior vs. posterior-anterior). Our approach leads to better predictive performance on external hospital data than both a standard baseline and previously proposed methods to handle confounding, and also suggests a method for identifying models that may rely on confounders.
Variants of uncertain significance (VUS) fundamentally limit the utility of genetic information in a clinical setting. The challenge of VUS is epitomized by
BRCA1
, a tumor suppressor gene integral to DNA repair and genomic stability. Germline
BRCA1
loss-of-function (LOF) variants predispose women to early-onset breast and ovarian cancers. Although
BRCA1
has been sequenced in millions of women, the risk associated with most newly observed variants cannot be definitively assigned. Data sharing attenuates this problem but it is unlikely to solve it, as most newly observed variants are exceedingly rare. In lieu of genetic evidence, experimental approaches can be used to functionally characterize VUS. However, to date, functional studies of
BRCA1
VUS have been conducted in a
post hoc
, piecemeal fashion. Here we employ saturation genome editing to assay 96.5% of all possible single nucleotide variants (SNVs) in 13 exons that encode functionally critical domains of BRCA1. Our assay measures cellular fitness in a haploid human cell line whose survival is dependent on intact
BRCA1
function. The resulting function scores for nearly 4,000 SNVs are bimodally distributed and almost perfectly concordant with established assessments of pathogenicity. Sequence-function maps enhanced by parallel measurements of variant effects on mRNA levels reveal mechanisms by which loss-of-function SNVs arise. Hundreds of missense SNVs critical for protein function are identified, as well as dozens of exonic and intronic SNVs that compromise
BRCA1
function by disrupting splicing or transcript stability. We predict that these function scores will be directly useful for the clinical interpretation of cancer risk based on
BRCA1
sequencing. Furthermore, we propose that this paradigm can be extended to overcome the challenge of VUS in other genes in which genetic variation is clinically actionable.