AI-Enabled Vessels Segmentation Model for Real-Time Laparoscopic Ultrasound Imaging

Ignas Kupcikevicius, Luca Boretto, Inger A. Grunbeck, Rahul Prasanna Kumar, Varatharajan Nainamalai, Mehdi Sadat Akhavi, Bjørn Edwin, Ole Jakob Elle

Keywords: real-time, ultrasound vessel segmentation, laparoscopic liver surgery, temporal triplet input, U-Net with ResNet-18 encoder, Artificial intelligence, sonogram detection, CLAHE

Abstract: Laparoscopic ultrasound (LUS) is essential for assessing the liver during laparoscopic liver resections. However, the interpretation of LUS images presents significant challenges due to the steep learning curve and image noise. In this study, we propose an enhanced U-Net-based neural network with a ResNet18 backbone specifically designed for real-time liver vessel segmentation of 2D LUS images. Our approach incorporates five preprocessing steps aimed at maximizing the training information extracted from the ultrasound sonogram region. The modified U-Net model achieved a Dice coefficient of 0.879, demonstrating real-time performance at 40 frames per second and enabling the development of advanced ultrasound-based surgical navigation solutions.

Serve As Reviewer: ~Rahul_Prasanna_Kumar1

Submission Number: 47