Abstract
Tuberculosis (TB) is a significant public health challenge worldwide. Early and accurate diagnosis is crucial for effective treatment and containment of the disease. This research work addresses the problem by proposing a multi-model classification method for identifying TB cases from chest X-rays with high accuracy. It utilizes a dataset created from real-time patient data collected from TB hospitals. Additionally, a comparative analysis of two deep learning models is conducted for the accurate detection of TB from chest X-ray images. The models were assessed based on accuracy, precision, recall, and F1-score, with the one unconventional model demonstrating superior performance. This paper discusses the potential reasons for the observed discrepancies, including differences in model architecture, data handling, and training processes. Our findings suggest that the integration of the softmax activation function into binary classification models can have a beneficial impact on training efficiency, leading to improved performance in medical image analysis for Tuberculosis detection. Although softmax is mathematically equivalent to sigmoid in binary tasks, our results indicate a potential advantage in utilizing softmax that warrants further investigation. To further enhance the robustness of our approach, future research will focus on incorporating additional datasets from diverse populations and exploring the integration of ensemble learning techniques, aiming to increase the generalizability and reliability of TB detection in chest X-ray images across varied demographic groups.
Keywords: Deep Learning, Image Analysis, Tuberculosis, X-Ray Analysis.