Advanced Lung Region Analysis and Anomaly Detection in CXR Images using Optimized VGG-16 Structure

The development of sophisticated frameworks for Chest X-Ray (CXR) analysis is required due to the growing need for precise and effective indicators in medical imaging. Therefore, this paper proposed a novel VGG-Residual Net architecture for the efficient diagnosis of Tuberculosis (TB). The preprocessing stage employs a Savitzky-Golay (S-G) filter to smooth noisy image data, ensuring high-quality inputs for analysis. Lung region segmentation is performed using the mean shift topology, which clusters pixels based on intensity and spatial proximity, effectively delineating anatomical structures. Local Binary Pattern (LBP) is used for feature extraction, capturing critical texture features that characterize spatial patterns in the lung region. The classification process leverages the VGG-Residual Net architecture to detect anomalies with higher accuracy. For the valuation of proposed topology, it is implemented in Python software and the outcomes ensure reliable and accurate diagnostic outcomes, even with complex datasets and limited labeled samples. It achieved good performance in terms of accuracy of 98.90%, precision of 97%, recall of 98%, and specificity of 98% compared to current advanced deep learning models.