Explainable deep-neural-network supported scheme for tuberculosis detection from chest radiographs

Abstract

Chest radiographs are examined in typical clinical settings by competent physicians for tuberculosis diagnosis. However, this procedure is time consuming and subjective. Due to the growing usage of machine learning techniques in applied sciences, researchers have begun applying comparable concepts to medical diagnostics, such as tuberculosis screening. In the period of extremely deep neural nets which comprised of hundreds of convolution layers for feature extraction, we create a shallow-CNN for screening of TB condition from Chest X-rays so that the model is able to offer appropriate interpretation for right diagnosis. The suggested model consists of four convolution-maxpooling layers with various hyperparameters that were optimized for optimal performance using a Bayesian optimization technique. The model was reported with a peak classification accuracy, F1-score, sensitivity and specificity of 0.95. In addition, the receiver operating characteristic (ROC) curve for the proposed shallow-CNN showed a peak area under the curve value of 0.976. Moreover, we have employed class activation maps (CAM) and Local Interpretable Model-agnostic Explanations (LIME), explainer systems for assessing the transparency and explainability of the model in comparison to a state-of-the-art pre-trained neural net such as the DenseNet.

Keywords: Class activation maps; Convolution neural network; Deep neural network; Explainable models; LIME explainer; Pre-trained model; Tuberculosis diagnosis.

Fig. 1

Data split for validating the CNN models used in this study

Fig. 2

Posterior-Anterior (PA) chest X-ray of the healthy and tuberculosis groups (TB)

Fig. 3

(a) Difference between Explainable and Conventional AI Models (b) Pipeline for the investigation strategy employed in this article to analyze tuberculosis for explainability

Fig. 4

Schematic of the proposed interpretable Shallow-CNN classification algorithm for tuberculosis

Fig. 5

Hyperparameter tuning process for selection of best configuration for the proposed Shallow-CNN TB classification model

Fig. 6

Training and validation process namely model accuracy and loss curves for the CNN models for tuberculosis classification studied in the proposed work. a AlexNet (b) Shallow-CNN (c) DenseNet

Fig. 7

Visualization of the convolutional layers' activation functions for the proposed shallow-CNN model's four layers (a-d) respectively

Fig. 8

Visualization of the Maxpooling layers' activation functions for the proposed shallow-CNN model's four layers

Fig. 9

CNN models classification performance for TB detection in terms of Receiver Operating Characteristic (ROC) curve and Confusion matrix (a) S-CNN (b) DenseNet (c) AlexNet

Fig. 10

The modified DenseNet model was evaluated using two explainable AI frameworks, CAM and LIME. CAM and LIME was used to generate heatmaps that show the dominant activation regions (red regions) for normal and TB images. The heatmaps showed that the model did not fully learn the features of the lung regions for both normal and TB images

Fig. 11

The proposed Shallow-Net model was evaluated using two explainable AI frameworks, CAM and LIME. The heatmaps showed that the model tried to learn the features of the lung regions specifically for both normal and TB images