Design of a Hybrid Multi-Scale Attention Framework for Pulmonary Nodule Detection Using Deep Learning

Pulmonary nodules represent some of the earliest signs of lung cancer; their early detection helps in improving the survival rates of patients. Despite the promising results obtained using deep learning in the detection of pulmonary nodules, several problems such as low sensitivity to small-sized nodules, high false positives, and poor model interpretability continue to exist. This paper suggests a novel framework referred to as Hybrid Multi-Scale Attention that combines multi-scale feature extraction, attention, feature fusion, and false-positive reduction. The hybrid multi-scale attention framework was developed with the gaps observed from the earlier review as the basis.