YOLOv8 Algorithm-aided Detection of Rib Fracture on Multiplane Reconstruction Images

Area of Science:

• Medical Imaging
• Artificial Intelligence in Radiology
• Diagnostic Accuracy

Background:

• Manual diagnosis of rib fractures on multiplane reconstruction (MPR) images is time-consuming.
• Current artificial intelligence (AI) models have limitations in rib fracture detection.
• There is a need for efficient and accurate AI-assisted diagnostic tools.

Purpose of the Study:

• To develop and evaluate a YOLOv8 algorithm-based detection model for identifying rib fractures on MPR images.
• To address the limitations of existing AI models and the labor-intensive nature of manual diagnosis.
• To assess the performance of the YOLOv8 model in terms of accuracy and speed.

Main Methods:

• Collected and curated a dataset of 624 CT-confirmed MPR images.
• Labeled images by a musculoskeletal radiologist.
• Trained and validated a YOLOv8n model on the dataset, calculating key performance metrics.

Main Results:

• The YOLOv8 model achieved high diagnostic accuracy: 96% sensitivity, 97% specificity, and 97% accuracy.
• The AI model interpreted 144 images in 2.02 seconds, significantly faster than a radiologist's 288 seconds.
• Demonstrated superior diagnostic speed compared to manual interpretation.

Conclusions:

• The YOLOv8 algorithm shows significant potential for rapid and accurate rib fracture diagnosis on MPR images.
• This AI tool can enhance clinical efficiency and reduce radiologist workload.
• Future research will focus on advanced feature learning and PACS integration for improved human-computer interaction.