Zone-Aware Pneumonia Classification Using Automated Lung Region Detection and Multi-Branch Feature Learning

Pneumonia is a major lung infection and remains one of the leading causes of death, especially among children under five and older adults. Early and accurate diagnosis often depends on chest X-rays, yet manual interpretation is challenging due to overlapping structures and low-contrast opacities. In this context, computer-based techniques have been proposed to assist radiologists in detecting pneumonia symptoms. Most existing techniques rely on deep learning models that extract features from the entire image to make classification decisions. While these approaches achieve high accuracy, they lack explainability, offering no clear information on where features are localized or how decisions are made. In this paper, we address two main limitations: (1) the localization of pneumonia-related features, and (2) the interpretability of deep features used for decision-making. This research presents a robust framework that integrates lung zone detection with multi-branch feature learning to classify pneumonia types and predict zone-specific findings. The proposed system consists of three phases. First, lung zone localization is performed to automatically focus on disease-relevant regions. Second, intra-zone CoT reasoning is introduced for zone-specific feature extraction, where lung zones detected by YOLOv8 are processed by ResNet-18 to capture local features. Third, attention-based inter-zone CoT fusion and classification is applied to predict pneumonia type and zone-level findings such as opacities and ground-glass patterns, aligning with radiology reporting standards. The proposed framework is evaluated on the Chest X-Ray Images (Pneumonia) dataset. It achieved an average accuracy of 94.4% and demonstrated notable improvements in detecting early-stage pneumonia. These results highlight the potential of the model as a decision-support tool for radiologists, enabling accurate diagnosis and standardized reporting in clinical practice.