Accurate and Robust Tuberculosis Detection Using Transferred Deep Learning Model

Abstract

Tuberculosis (TB) is an infectious illness that is extremely dangerous to people's health. Early detection can help with therapy and reduce disease spread. Automatic TB detection is essential when there is a lack of diagnostic resources and a requirement for rapid detection. Automatic TB detection is a computer-based diagnostic system that may generate a rapid and precise diagnostic report as an output with less error than a human diagnosis report. This will give the ability to produce more effective mass screening with less resources. The major goal of this research paper is to create an effective computer-aided detection system that will help clinicians make more informed Tuberculosis diagnoses. The paper adopts a deep learning model to detect Tuberculosis infection. To lessen the need for data and hyperparameter adjustment, transfer learning was used. The transfer learning model was VGG-19, and a total of 4200 images were used to evaluate the model's performance. The test results show an accuracy 99.444%, which is achieved, and the recall is 97.143%. This suggests that the proposed model is accurate and robust and can be adopted as a means for TB detection in the future through real-time implementation.