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

Abstract

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

Artificial Intelligence-Based Diagnosis of Cardiac and Related Diseases

“…Thus, the output of X-RayNet is the four masks for each individual class output. To evaluate the segmentation performance by the proposed X-RayNet, the accuracy (Acc); mean intersection of union (mIOU), which is also referred as the Jaccard index (J); and dice coefficient (D) were measured, which were similarly utilized by [1,12,47] to evaluate and compare the JSRT dataset with other methods. The formulas for J and D are given by Equations (3)- (5).…”

“…Cardiomegaly can be assessed by the cardiothoracic ratio (CTR), which is measured manually by medical experts using the boundaries of the lungs and heart in CXRs [8]. Several studies have evaluated segmentation of the chest anatomy to estimate the CTR for cardiomegaly and related diseases [9][10][11][12][13]. To obtain advancement in diagnosis, automated systems are required to aid the medical specialist and overcome the diagnostic burden [2,3].…”

“…Solovyev et al presented a novel method for the estimation of the CTR. They utilized the feature pyramid network (FPN) decoder with the change of batch normalization for instance normalization and to incorporate dropout in the network [12]. Oliveira et al proposed transfer learning-based semantic segmentation for multiclass chest anatomy segmentation.…”

Automated analysis of rabbit knee calcified cartilage morphology using micro-computed tomography and deep learning segmentation

“…The fully automatic CC segmentation was conducted using a deep learning pipeline inspired by Solovyev et al (40) on Python 3.7. The pipeline was built using in-house developed Collagen-framework (https://github.com/MIPT-Oulu/Collagen).…”

“…Notably, such approach works efficiently across domains beyond natural images (36,37) . For example, transfer learning from deep residual networks (38) has been used to classify pulmonary nodules from CT images (39) , or segment the lungs in chest X-rays (40) .…”