Bayesian Feature Pyramid Networks for Automatic Multi-Label Segmentation of Chest X-rays and Assessment of Cardio-Thoratic Ratio

Cardiothoratic ratio (CTR) estimated from chest radiographs is a marker\nindicative of cardiomegaly, the presence of which is in the criteria for heart\nfailure diagnosis. Existing methods for automatic assessment of CTR are driven\nby Deep Learning-based segmentation. However, these techniques produce only\npoint estimates of CTR but clinical decision making typically assumes the\nuncertainty. In this paper, we propose a novel method for chest X-ray\nsegmentation and CTR assessment in an automatic manner. In contrast to the\nprevious art, we, for the first time, propose to estimate CTR with uncertainty\nbounds. Our method is based on Deep Convolutional Neural Network with Feature\nPyramid Network (FPN) decoder. We propose two modifications of FPN: replace the\nbatch normalization with instance normalization and inject the dropout which\nallows to obtain the Monte-Carlo estimates of the segmentation maps at test\ntime. Finally, using the predicted segmentation mask samples, we estimate CTR\nwith uncertainty. In our experiments we demonstrate that the proposed method\ngeneralizes well to three different test sets. Finally, we make the annotations\nproduced by two radiologists for all our datasets publicly available.\n