YOLOv8-BCD: a real-time deep learning framework for pulmonary nodule detection in computed tomography imaging

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Computer Vision

Background:

• Lung cancer is a leading cause of cancer mortality globally.
• Early detection of pulmonary nodules via computed tomography (CT) is crucial for lung cancer screening.
• Traditional nodule detection methods face limitations in accuracy and efficiency.

Purpose of the Study:

• To develop an advanced deep learning framework for high-precision and rapid pulmonary nodule identification in CT images.
• To facilitate earlier and more accurate lung cancer diagnosis.

Main Methods:

• Proposed an improved deep learning framework, YOLOv8-BCD, integrating YOLOv8 with BiFormer attention, Content-Aware ReAssembly of Features (CARAFE), and Depth-wise Over-Parameterized Depth-wise Convolution (DO-DConv).
• Employed Super-Resolution Generative Adversarial Network (SRGAN) for image enhancement to address low resolution, noise, and artifacts in CT images.
• Incorporated BiFormer attention in the backbone for enhanced feature extraction, especially for small nodules, and utilized CARAFE and DO-DConv in the head for optimized feature fusion and reduced computational complexity.

Main Results:

• On the LUNA16 dataset, YOLOv8-BCD achieved 86.4% detection accuracy and 88.3% mAP_0.5, outperforming YOLOv8.
• External validation on the TianChi dataset showed an mAP_0.5 of 83.8% and mAP_0.5-0.95 of 43.9% with a high inference speed of 98 FPS.
• The model demonstrated superior performance in accuracy and mean average precision (mAP) compared to the baseline YOLOv8.

Conclusions:

• The YOLOv8-BCD model significantly aids clinicians by reducing interpretation time and improving diagnostic accuracy.
• The framework helps minimize missed diagnoses, ultimately enhancing patient outcomes in lung cancer screening.
• This deep learning approach offers a promising tool for more effective early lung cancer detection.