Development of a Two Stage Pipeline for Robust Pulmonary Nodule Detection in Consecutive CT Slices
Year: 2026
Type: article
Abstract: Background/Objectives: Current pulmonary nodule detection systems often neglect the spatiotemporal continuity inherent in computed tomography (CT) sequences within a single scan. They treat individual slices as independent images and lack mechanisms to recover missed detections. As a result, nodules missed in individual slices cannot be recovered. This study aims to develop a detection–tracking co-design framework that improves both sensitivity and specificity for pulmonary nodule detection. Methods: We propose a detection-tracking co-design framework. It couples an enhanced YOLOX detector with a Kalman filter-based tracker in Stage I, to associate nodule candidates across consecutive slices and recover missed detections. Stage II employs cross-slice Maximum Intensity Projection and a lightweight ResNet-50 classifier, to reduce false positives through morphological feature discrimination. Results: Systematic ablation studies on the Lung Image Database Consortium and Image Database Resource Initiative (LIDC-IDRI) dataset using the criteria of the Lung Nodule Analysis 2016 (LUNA16) dataset demonstrate the complementary contributions of each component. The complete framework achieves a Competition Performance Metric (CPM) score of 0.9344 and exhibits particular strength in medium-to-high sensitivity regions. Conclusions: The proposed detection–tracking co-design provides an interpretable paradigm that balances high sensitivity and improved specificity, offering a promising solution for computer-aided diagnosis of pulmonary nodules.
Source: Biomedicines
Language: English
Topic: AI in cancer detection
Subfield: Artificial Intelligence
Field: Computer Science
Domain: Physical Sciences
Open Access status: gold
Estimated APC paid: $3,137
Awards: G2878097952, G5032662597, G5525507772
OpenAlex ID: W7204436897