Detection of Tuberculosis Based on Hybridized Pre-Processing Deep Learning Method

The disease, tuberculosis (TB) is a serious health concern as it primarily affects the lungs and can lead to fatalities. However, early detection and treatment can cure the disease. One potential method for detecting TB is using Computer Aided Diagnosis (CAD) systems, which can analyze Chest X-Ray Images (CXR) for signs of TB. This paper proposes a new approach for improving the performance of CAD systems by using a hybrid pre-processing method for Convolutional Neural Network (CNN) models. The goal of the research is to enhance the accuracy and Area Under Curve (AUC) of detection for TB in CXR images by combining two different pre-processing methods and multi-classifying different manifestations of the disease. The hypothesis is that this approach will result in more accurate detection of TB in CXR images. To achieve this, this research used augmentation and segmentation techniques to pre-process the CXR images before feeding them into a pre-trained CNN model for classification. The VGG16 model managed to achieve an AUC of 0.935, an accuracy of 90% and a 0.8975 F1-score with the proposed pre-processing method.