Extraction of the liver falciform ligament and ridgeline in preoperative point clouds based on PointNet++ and transformer models

Liver falciform ligament and ridgeline are important anatomical features in preoperative liver point cloud. This paper aims to extract these features. We propose a liver falciform ligament and ridgeline extraction method based on PointNet++ and Transformer. The Transformer module integrates a self-attention layer, linear projections that reduce the dimensionality and accelerates the processing, and residual connection, which can extract falciform ligament and ridgeline by global and local features of the liver point cloud. Our method can extract liver falciform ligament and ridgeline. The mAcc and mIoU have 2.62% and 1.46% improvement compared to PointNet++. The prediction of the liver falciform ligament has a lower failure rate than PointNet++, and the average distance error for the nearest point of the falciform ligament is 7.64 mm. The PointNet++-Transformer-based liver falciform ligament and ridgeline extraction method has high accuracy and low failure rate, as well as robustness.