Computer-Aided System for the Detection of Multicategory Pulmonary Tuberculosis in Radiographs

Abstract

The early screening and diagnosis of tuberculosis plays an important role in the control and treatment of tuberculosis infections. In this paper, an integrated computer-aided system based on deep learning is proposed for the detection of multiple categories of tuberculosis lesions in chest radiographs. In this system, the fully convolutional neural network method is used to segment the lung area from the entire chest radiograph for pulmonary tuberculosis detection. Different from the previous analysis of the whole chest radiograph, we focus on the specific tuberculosis lesion areas for the analysis and propose the first multicategory tuberculosis lesion detection method. In it, a learning scalable pyramid structure is introduced into the Faster Region-based Convolutional Network (Faster RCNN), which effectively improves the detection of small-area lesions, mines indistinguishable samples during the training process, and uses reinforcement learning to reduce the detection of false-positive lesions. To compare our method with the current tuberculosis detection system, we propose a classification rule for whole chest X-rays using a multicategory tuberculosis lesion detection model and achieve good performance on two public datasets (Montgomery: AUC = 0.977 and accuracy = 0.926; Shenzhen: AUC = 0.941 and accuracy = 0.902). Our proposed computer-aided system is superior to current systems that can be used to assist radiologists in diagnoses and public health providers in screening for tuberculosis in areas where tuberculosis is endemic.

Figure 1

The distributions and sizes of the six tuberculosis lesions, with the width as the abscissa and the height as the ordinate.

Figure 2

The relationship between the frequency of occurrence of each grey-scale value and the grey-scale value in the tuberculosis lesion image, with the grey-scale value as the abscissa and the frequency as the ordinate.

Figure 3

Comprehensive view of the computer-aided system for tuberculosis diagnosis using chest radiography.

Figure 4

Segmentation model architecture.

Figure 5

The learning scalable feature pyramid architecture in the Faster RCNN network. The architecture of the discovered 7-merging-cell pyramid network in the NAS-FPN includes 5 input layers (Navy blue) and 5 output feature layers (Light blue). GP and R-C-B stand for Global Pooling and ReLU-Conv-BatchNorm, respectively.

Figure 6

Segmentation results: (a) JSRT and (b) Montgomery. The columns are the original image, segmented mask, ground truth, and difference of the segmented mask from the ground truth from left to right.

Figure 7

The red boxes are the ground truths, and the green, blue, and white boxes represent the results of our proposed method. The lesion categories: (a) exudation, (b) nodules, (c) calcification, (d) miliary tuberculosis, (e) encapsulated pleural effusion, and (f) free pleural effusion.

Figure 8

The Performance of the proposed computer-aided system on whole chest X-rays on local dataset (MyDataset) and two public datasets, the Montgomery dataset (MC), and the Shenzhen dataset (ShenZhen). (a) ROC, (b) accuracy, (c) false positive rate, and (d) true positive rate.