Inguinal hernia repair is one of the most frequently conducted surgical procedures worldwide. Laparoscopic inguinal hernia repair is considered to be technically challenging. Artificial intelligence technology has made significant progress in medical imaging, but its application in laparoscopic surgery has not been widely carried out. Our aim is to detect vas deferens images in laparoscopic inguinal hernial repair using the convolutional neural network (CNN) and help surgeons to identify the vas deferens in time. We collected surgery videos from 35 patients with inguinal hernia who underwent laparoscopic hernia repair. We classified and labeled the images of the vas deferens and used the CNN to learn the image features. Totally, 2,600 images (26 patients) were labeled for training and validating the neural network and 1,200 images (6 patients) and 6 short video clips (3 patients) for testing. We adjusted the model parameters and tested the performance of the model under different confidence levels and IoU and used the chi-square to analyze the statistical difference in the video test dataset. We evaluated the model performance by calculating the true positive rate (TPR), true negative rate (TNR), accuracy (ACC), positive predictive value (PPV), and F1-score at different confidence levels of 0.1 to 0.9. In confidence level 0.4, the results were TPR 90.61%, TNR 98.67%, PPV 98.57%, ACC 94.61%, and F1 94.42%, respectively. The average precision (AP) was 92.38% at IoU 0.3. In the video test dataset, the average values of TPR and TNR were 90.11% and 95.76%, respectively, and there was no significant difference among the patients. The results suggest that the CNN can quickly and accurately identify and label vas deferens images in laparoscopic inguinal hernia repair.
Copyright © 2021 Peng Cui et al.