Research on lung nodule detection in X-ray plain films based on improved YOLOv12 model

Area of Science:

• Medical Imaging
• Computer Vision
• Artificial Intelligence

Background:

• Accurate lung nodule detection in chest X-rays is crucial for early diagnosis of lung diseases.
• Existing deep learning models face challenges in balancing detection accuracy and computational efficiency.

Purpose of the Study:

• To develop an improved YOLOv12-based framework for enhanced automatic lung nodule detection.
• To optimize detection performance and reduce computational complexity in chest X-ray analysis.

Main Methods:

• Integration of Space-to-Depth Convolution (SPDConv) for spatial information preservation.
• Implementation of a Dynamic Upsampling module (DySample) for improved multi-scale feature fusion.
• Inclusion of a lightweight feature aggregation module (VoVGSCSP) to strengthen feature representation and reduce redundancy.

Main Results:

• The proposed YOLOv12-DSV model achieved a higher mAP50 (0.735) and mAP50-95 (0.426) compared to the original YOLOv12.
• Reduced model parameters from 2.52M to 2.21M and FLOPs from 6.0G to 5.2G.
• Increased inference speed from 97.6 FPS to 107.8 FPS.

Conclusions:

• The YOLOv12-DSV model demonstrates superior lung nodule detection accuracy in chest X-rays.
• The optimized model offers a better trade-off between detection performance and computational cost.
• This framework provides a more efficient solution for automated lung nodule localization.