Optimal Deep Transfer Learning Model for Automated Tuberculosis Classification on Chest Radiographs

Tuberculosis (TB) is the fifth leading cause of mortality rates across the world, adding nearly 10 million new cases and 1.5 million deaths annually. As TB caused by the bacteria that mostly affect the lungs is prevented and cured, the World Health Organization (WHO) reported a systematic and broad screening for eradicating the disease. Despite its interpretational difficulty and low specificity, poster anterior (PA) chest radiography becomes one of the preferred TB screening techniques. TB is majorly a disease in poor nations; thus, medical practitioners trained to interpret such CXRs were rare. Numerous computer-aided diagnosis (CAD) researches which deal with CXR abnormalities do not give more attention to other diseases (i.e., non-TB). This article devises an Optimal Deep Transfer Learning Model for Automated Tuberculosis Classification (ODTLATC) model. The presented ODTLATC model majorly concentrates on the identification of TB on chest radiographs. To attain this, the ODTLATC model follows a three-stage process such as pre-processing, feature extraction, and classification. At the initial stage, the ODTLATC model employs Weiner filtering (WF) approach for image denoising process. For feature extraction, deep convolutional neural network based residual network (ResNet50) model is utilized. At last, whale optimization algorithm (WOA) with bidirectional recurrent neural network (BiRNN) model is exploited for TB classification purposes. To demonstrate the better performance of the ODTLATC model, a extensive variety of simulations are conducted and the outcomes were inspected on chest radiographs. The comparative study reported the improved performance of the ODTLATC model over other DL models.