Preoperative to Intraoperative Laparoscopy Fusion

Augmented reality (AR) in laparoscopic liver surgery needs key landmark detection in intraoperative 2D laparoscopic images and its registration with the preoperative 3D model created from CT/MRI data. Such AR techniques are vital to surgeons as they enable precise tumor localisation for surgical removal. A full resection of targeted tumor minimises the risk of recurrence. However, the task of automatic anatomical curve segmentation (considered as landmarks), and its registration to 3D models is a non-trivial and complex task. Most developed methods in this domain are built around traditional methodologies in computer vision. This challenge is designed to challenge participants to deploy machine learning methods for two tasks - a) task I: segmentation of key anatomical curves from laparoscopic video images and 3D model, and b) task 2: matching these segmented curves to the 3D liver model from volumetric data (CT/MR). Thus the challenge is aimed at segmenting anatomical curves such as ridges, liver contours and midline of ligament (supervised and semi-supervised), and 2D-3D registration problem (semi-supervised and unsupervised) between the segmented landmarks with the provided dense 3D point cloud of liver. Here, we will assess the quality of registration algorithms using widely used target registration errors while liver landmark segmentation will be evaluated-based on F1-score