Artificial Intelligence-Assisted Liver Ultrasound Training: Image Recognition and Real-Time Feedback System Based on YOLO and DeepSeek.

RATIONALE AND

OBJECTIVES

This study aimed to develop and evaluate an artificial intelligence‑assisted liver ultrasound training system that combines the YOLOv11n object detection algorithm and the DeepSeek large language model for ultrasound education.

MATERIALS AND

METHODS

A parallel controlled study was performed with 20 students randomly assigned to either the traditional instruction group (n = 10) or the AI‑assisted training group (n = 10). The YOLOv11n model was trained using 326 liver ultrasound images. Model performance metrics including precision, recall, mAP@0.5 and mAP@0.5:0.95 were calculated. System efficiency indicators, section recognition accuracy, evaluation consistency and post‑training examination scores were compared between groups. Grad‑CAM++ heatmaps were applied to visualize the model's attention on anatomical features.

RESULTS

The YOLOv11n model yielded a precision of 0.984, a recall of 1.00, an mAP@0.5 of 0.995, and an mAP@0.5:0.95 of 0.7099. Compared with traditional training, the AI‑assisted system reduced daily report processing time by 96.3% (from 486 to 18 min) and decreased feedback latency by 95.3% (from 44.8 to 2.1 min). Section recognition accuracy increased from 86.0% to 96.0%, evaluation consistency rose from 90.0% to 99.6%, and post‑training examination scores improved from 87.9 to 96.2 points. Grad‑CAM++ heatmaps verified that the model focused on anatomically meaningful regions.

CONCLUSION

Integration of computer vision and large language models can improve teaching efficiency and learning performance in liver ultrasound training.