A new method for false-positive reduction in detection of lung nodules in CT images

Abstract

This paper proposes a novel approach for false-positive reduction in lung nodule detection based on structure relationship analysis between nodule candidate and vessel, and the modified surface normal overlap descriptor. On one hand, a large number of false nodules attached to vessels can be removed by analyzing the relationship between nodule candidates and their attached tissues. On the other hand, Low-contrast nonsolid nodules are discriminated from the candidates with modified surface normal overlap descriptor. The proposed method has been trained and validated on a clinical dataset of 90 thoracic CT scans using a low dose levels that contain 90 nodules (62 solid nodules, 25 ground-glass opacity nodules and 3 mixed nodules) determined by a ground truth reading process.