A data-constrained approach for occupational silicosis detection on chest X-rays with few-shot learning

Abstract

Occupational silicosis is a serious lung disease caused by long-term exposure to silica dust, mainly affecting workers in industries such as mining and construction. Diagnosis of silicosis is challenging due to subtle disease manifestations on chest X-rays (CXRs) and limited labeled medical data. Traditional deep learning models, such as Convolutional Neural Networks (CNNs), often require large datasets, which are often heavily expensive and time-consumed for collection and annotation, yet useful for specialized medical applications. To address these challenges, we present the use of Few-Shot Learning (FSL) to enable accurate the detection of occupational silicosis with a minimal number of labeled examples. Our experimental results demonstrate that the FSL-based model achieves 84.4% accuracy and 46.0% mIoU in the 1-shot setting and 89.52% accuracy with 47.89% mIoU in the 4-shot setting. These findings highlight the potential of FSL to improve diagnostic accuracy in data-limited environments, making it a viable solution for improving medical image analysis in resource-constrained settings