Enhanced object detection in pediatric bronchoscopy images using YOLO-based algorithms with CBAM attention mechanism

Area of Science:

• Medical Imaging
• Computer Vision
• Respiratory Medicine

Background:

• Bronchoscopy is crucial for diagnosing respiratory conditions like infections and tumors.
• Visualizing bronchial anatomy in bronchoscopy can be challenging due to artifacts and complexity.
• Accurate identification of bronchial structures is vital for effective diagnosis and treatment.

Purpose of the Study:

• To improve the accuracy of object detection in bronchoscopy images.
• To integrate a YOLO-based object detection algorithm with a CBAM attention mechanism.
• To enhance the identification of key anatomical features within the bronchial tree.

Main Methods:

• Implemented the CBAM attention module within YOLO-V7 and YOLO-V8 object detection architectures.
• Compared the performance of YOLO-V5, YOLO-V7, and YOLO-V8 on a dataset of bronchoscopy images.
• Evaluated the object detection performance using metrics such as mean Average Precision (mAP) and Intersection over Union (IoU).

Main Results:

• The YOLO-V8 model achieved an mAP of 87.09% at an IoU threshold of 0.5.
• Integrating the CBAM attention module into YOLO-V8 significantly improved performance, achieving an mAP of 88.27%.
• The proposed method demonstrated enhanced accuracy and reliability in detecting objects within bronchoscopy images.

Conclusions:

• Incorporating a CBAM attention mechanism with YOLO algorithms notably improves object detection in bronchoscopy.
• This approach offers significant potential for enhancing medical imaging analysis in respiratory diagnostics.
• The study provides valuable insights into optimizing attention mechanisms for object detection in clinical settings.