The increasing prevalence of respiratory diseases, including the recent COVID-19 pandemic, underscores the need for advanced diagnostic tools that can enhance early detection and management strategies. This paper introduces a refined approach using the U-Net neural network architecture for the analysis of chest images, aimed at improving the outcomes in the diagnosis of respiratory ailments. The U-Net model, originally designed for biomedical image segmentation, is adapted here to exploit its robust feature extraction capabilities that are critical for handling the complexities associated with chest radiographs, such as variability in organ size, shape, and pathological manifestations. Through a series of transpose convolutions, pooling, and concatenation layers, the model captures intricate details across multiple spatial scales, significantly improving the accuracy of segmentation and detection tasks. This study not only explores the architecture’s efficiency in segmenting and classifying these images but also enhances it with explainable AI techniques to ensure transparency and reliability in automated decision-making.