The Impact of Image Pre-processing for Tuberculosis Prediction System Based on Chest X-ray Images

With the rapid development of automated detection system using deep learning techniques on Chest X-ray (CXR) image datasets to the subjective assessment performed by healthcare professionals. Preprocessing is critical in medical image analysis as it helps highlight important anatomical features while suppressing irrelevant information, thus enabling the model to focus on meaningful patterns. In this paper, we investigate the impact of image preprocessing techniques on the performance of a tuberculosis prediction system based on CXR images using a deep learning approach. We used the “Tuberculosis Chest X-rays (Shenzhen)” dataset, which contains 1,344 CXR images (672 TB cases and 672 normal cases). We propose a five-step preprocessing pipeline consisting of resizing, heavy sharpen filtering, CLAHE (Contrast Limited Adaptive Histogram Equalization), horizontal flip augmentation, and data normalization. The findings indicate that the model utilising preprocessing markedly surpasses the one lacking it, attaining an accuracy, precision, recall, and F1-score of 71%, in contrast to 51%, 50%, 50%, and 36% without preprocessing, respectively. This study enhances the existing research on the application of deep learning in medical diagnostics and emphasises the significance of preprocessing for attaining dependable, high-performance systems.