Hybrid Cascaded Neural Network for Liver Lesion Segmentation

TL;DR

This paper introduces a hybrid cascaded neural network combining 2D and 3D CNNs to improve liver lesion segmentation accuracy, especially for small lesions, achieving top performance on the LiTS challenge.

Contribution

It presents a novel cascaded system that integrates 2D and 3D CNNs for more effective hepatic lesion segmentation, addressing limitations of previous single-model approaches.

Findings

Achieved a Dice score of 68.1% on LiTS challenge

Outperformed all non pre-trained models in the challenge

Identified annotation issues through cross-validation

Abstract

Automatic liver lesion segmentation is a challenging task while having a significant impact on assisting medical professionals in the designing of effective treatment and planning proper care. In this paper we propose a cascaded system that combines both 2D and 3D convolutional neural networks to effectively segment hepatic lesions. Our 2D network operates on a slice by slice basis to segment the liver and larger tumors, while we use a 3D network to detect small lesions that are often missed in a 2D segmentation design. We employ this algorithm on the LiTS challenge obtaining a Dice score per case of 68.1%, which performs the best among all non pre-trained models and the second best among published methods. We also perform two-fold cross-validation to reveal the over- and under-segmentation issues in the LiTS annotations.