Optimized tuberculosis image classification in chest X-rays using transfer learning and hyper parameters

Timely and accurate detection of tuberculosis (TB) is crucial, as it remains a significant global health concern. In 2023, an estimated 10 million new TB cases were recorded, with 9.8 million affecting adults and 0.2 million children. Tragically, TB was responsible for approximately 1.5 million deaths worldwide, highlighting the need for advanced diagnostic tools. This research focuses on developing a reliable TB classification model using chest X-ray images, employing deep learning methods such as Convolutional Neural Networks (CNNs). The methodology integrates transfer learning and hyper parameter tuning for improved performance. The dataset comprises 4200 images, including 3500 normal cases and 700 TB-positive cases, sourced from Kaggle's "Tuberculosis (TB) Chest X-ray Database." Leveraging the pre-trained VGG19 architecture, the model utilizes transfer learning techniques, task-specific layer adjustments, regularization methods, and layer freezing for precise classification. The proposed approach achieves an impressive accuracy of 98%, with detailed analysis addressing potential limitations despite its strong performance.