Tuberculosis detection in chest radiograph using convolutional neural network architecture and explainable artificial intelligence

Abstract

In most regions of the world, tuberculosis (TB) is classified as a malignant infectious disease that can be fatal. Using advanced tools and technology, automatic analysis and classification of chest X-rays (CXRs) into TB and non-TB can be a reliable alternative to the subjective assessment performed by healthcare professionals. Thus, in the study, we propose an automatic TB detection system using advanced deep learning (DL) models. A significant portion of a CXR image is dark, providing no information for diagnosis and potentially confusing DL models. Therefore, in the proposed system, we use sophisticated segmentation networks to extract the region of interest from multimedia CXRs. Then, segmented images are fed into the DL models. For the subjective assessment, we use explainable artificial intelligence to visualize TB-infected parts of the lung. We use different convolutional neural network (CNN) models in our experiments and compare their classification performance using three publicly available CXR datasets. EfficientNetB3, one of the CNN models, achieves the highest accuracy of 99.1%, with a receiver operating characteristic of 99.9%, and an average accuracy of 98.7%. Experiment results confirm that using segmented lung CXR images produces better performance than does using raw lung CXR images.

Keywords: Chest X-Ray; Convolution neural networks; Deep learning; Image segmentation; Tuberculosis detection.

Fig. 1

Chest X-ray view

Fig. 2

Different types of diseases in CXR

Fig. 3

Steps for lung segmentation: a original image, b lung mask, c output segmentation, and d segmented region of interest

Fig. 4

U-Net architecture

Fig. 5

Block diagram of the proposed TB detection system

Fig. 6

Image negated from Belarus dataset

Fig. 7

Data distribution in combined dataset experiments for normal and segmented CXR images

Fig. 8

CXR images after augmentation using nine different angles: a 0°, b 45°, c 90°, d 135°, e 180°, f 225°, g 270°, h 315°, and i 360°

Fig. 9

U-Net output segmentation results for two sample lung CXR images. a shows the original images. b shows the results of U-Net. c show the results after ROI extraction

Fig. 10

Model loss for augmented segmented CXR images

Fig. 11

Comparison between five CNN models using ROC-AUC

Fig. 12

Confusion matrix for efficientNetB3 for segmented augmented CXR images in the combined dataset. Left: a confusion matrix for the best accuracy using cross-validation; right: a confusion matrix using average rounds

Fig. 13

t-SNE for normal CXR in the combined dataset

Fig. 14

t-SNE for segmented CXR images in the combined dataset

Fig. 15

t-SNE for normal CXR images in the combined dataset after applying augmentation

Fig. 16

t-SNE for segmented CXR images in the combined dataset after applying augmentation

Fig. 17

Grad-CAM visualization of classified TB in raw CXR images (without augmentation and segmentation: a ResNet50; b Xception; c MobileNet; d InceptionResNetV2; and e EfficientNetB3

Fig. 18

Grad-CAM visualization of classified TB in raw CXR images with augmentation without segmentation: a ResNet50; b Xception; c MobileNet; d InceptionResNetV2; and e EfficientNetB3

Fig. 19

Grad-CAM visualization of classified TB in segmented CXR images without augmentation: a ResNet50; b Xception; c MobileNet; d InceptionResNetV2; and e EfficientNetB3

Fig. 20

Grad-CAM visualization of classified TB in segmented CXR images with augmentation: a ResNet50; b Xception; c MobileNet; d InceptionResNetV2; and e EfficientNetB3