A Low-Cost Rapid X-ray Image Screening System: Base on Convolutional Neural Networks

Detecting pulmonary nodules in chest X-rays (CXRs) is essential for early lung disease screening. This study developed a lightweight Convolutional Neural Network (CNN)-based Computer-Aided Detection (CAD) system leveraging YOLOv2 and YOLOv8 to operate efficiently in resource-constrained medical environments. The models were trained and validated using datasets from the NIH and Vietnamese medical centers, with annotations verified by radiologists from a southern Taiwan medical center. The proposed system demonstrated high accuracy and efficiency, effectively detecting pulmonary nodules, including nodular opacities from both abnormalities and overlapping anatomical structures. YOLOv2 showed particular strength in handling complex regions, such as retrocardiac and clavicular overlaps. Its lightweight design makes it suitable for primary care facilities, remote clinics, and as an auxiliary tool in larger hospitals to enhance diagnostic efficiency. Future work will expand datasets, optimize model structures, and conduct clinical trials to validate real-world applications. Efforts will also focus on model compression and hardware acceleration to further reduce computational costs. While this study targets CXRs, the system’s architecture can be adapted for other medical imaging modalities, ensuring broad applicability and clinical relevance.