Analysis of Tuberculosis Segmentation on Chest X-Rays Using the Method of Convolutional Neural Network “Attention”

This study presents an experimental study using an advanced Attention U-Net deep learning model to improve tuberculosis (TB) segmentation in chest X-ray images. The goal of this study is to use convolutional neural networks (CNNs) with integrated attentionmechanisms improve the accuracy of medical image analysis. In the experiment, two different datasets were combined and augmented, resulting in a complete set of 800 images. These images were used to thoroughly train and validate the model's performance. The methodical approach allowed the Attention U-Net model to effectively highlight meaningful features while suppressing extraneous background noise. This ability is important for achieving accuracy in medical diagnosis. The empirical results were convincing: the model demonstrated outstanding segmentation abilities, as evidenced by high metrics: accuracy reached 0.9824, Dice score was 0.9767, and IOU was 0.9778. The integration of attention mechanisms into the CNN architecture has improved accuracy, leading to better detection and diagnosis of lung diseases. This study illustrates how the application of advanced machine learning techniques can help radiologists in image analysis.