Crop and Segment: Efficient Whole body pan-cancer segmentation

Yinyin Luo, Yue Liu, Wenbin Liu, Jingheng Dai, Xunliang Xiao, Gang Fang

Keywords: Semi-supervised · Deep learning · Tumour segmentation

TL;DR: We explored the application of nnUNet for whole-body tumor segmentation in CT scans, Proposing a more precise cropping strategy and introducing an organ-interference segmentation approach to effectively enhance segmentation efficiency.

Abstract:

Despite significant advancements in deep learning models for
medical segmentation, the detection and segmentation of tumors, particularly through whole-body scans, remain challenging. To address this
issue, we explored the application of nnUNet for whole-body tumor segmentation in CT scans, proposing a more precise cropping strategy and
introducing an organ-interference segmentation approach to effectively
enhance segmentation efficiency. Experiments on the MICCAI FLARE
2024 dataset demonstrated significant improvements in both segmentation accuracy and efficiency. Our method achieved an average organ Dice
Similarity Coefficient (DSC) of 10.47% and a Normalized Surface Dice
(NSD) of 7.98% on the public validation set. In the FLARE 2024 Task
1 online validation, the method achieved an average organ Dice Similarity Coefficient (DSC) of 17.08%, a Normalized Surface Dice (NSD) of
7.42% and the average running time and area under GPU memory-time
curve were 19.89s and 45688 MB, respectively. The code is available at
https://github.com/lay-john/FLARE24-Task1.

Submission Number: 9