A deep-learning pipeline to assist portal vein identification during laparoscopic ultrasound scanning for anatomical liver resection in real-time

Objective Intraoperative ultrasound (IOUS) is frequently used for real-time guidance during anatomical liver resection, which is considered to be the gold standard treatment for early-stage hepatocellular carcinoma (HCC), but manual hepatic vessel annotation continues to be challenging due to the complexity and noise in ultrasound images. Materials and Methods With an emphasis on anatomical liver resection, our research examines the practicality of using the YOLOv5 deep-learning algorithm for real-time identification of hepatic vessels in ultrasound images. The model was rained, tested, and validated using a dataset of 100 intraoperative laparoscopic ultrasound images. Also, real-time hepatic vessel recognition was performed intraoperatively in 20 patients. Results The results showed engrossing performances, with the model achieving a precision of 1 when confidence reached 0.764 in the P-curve analysis. The R-curve analysis revealed that recall dropped to 0 at a confidence threshold of 1.00. When the Intersection over Union (IoU) was set to 0.5, the mean Average Precision (mAP) was 0.941. Additionally, the optimal balance between precision and recall was observed at a confidence level of 0.202, where the F1 score reached 0.58. Conclusion These findings demonstrate that the YOLOv5 model has a major potential to enhance surgical precision, reduce duration of surgery, and reduce surgical riscks. This study highlights the potential of AI models in improving outcomes in liver surgeries and suggests broader applications in other surgical domains.