To develop a lightweight yet accurate detection framework for distinguishing adenocarcinoma in situ (AIS) from minimally invasive adenocarcinoma (MIA) in pulmonary CT imaging.
Approach:
FAD-YOLO Framework: The framework includes three improvements: A2C2f-FRFN module for feature refinement, DySample for dynamic upsampling, and TADDH for task-aligned detection.
Key Findings:
FAD-YOLO achieved a precision of 93.8% and a recall of 93.4% on an internal test set of 317 images.
The model reduced parameters by 18.7% compared to the baseline while increasing GFLOPs by 12.5%.
FAD-YOLO outperformed YOLOv5n, YOLOv8n, YOLO11n, and RT-DETR-R18, and surpassed RT-DETR-R50 on all accuracy metrics despite using significantly fewer parameters.
On an external test set, FAD-YOLO achieved mAP@50 of 91.7% and mAP@50–95 of 68.2% without fine-tuning.
Interpretation:
Limitations:
The study does not provide prospective validation of the FAD-YOLO framework.
The performance may vary across different datasets and clinical environments.
