Yu Cenji, Gay Skylar S, Gupta Aashish C, Martin-Paulpeter Rachael M, Ludmir Ethan B, Zhao Yao, Duryea Jack, Chen Xinru, Cardenas Carlos E, Yang Jinzhong, Koong Albert C, Netherton Tucker J, Rhee Dong Joo, Court Laurence E
The University of Texas MD Anderson Cancer Center UTHealth Graduate School of Biomedical Sciences (GSBS), 6767 Bertner Avenue, Houston, TX 77030, USA.
Department of Radiation Physics, The University of Texas MD Anderson Cancer Center, 1400 Pressler Street, Houston, TX 77030, USA.
Phys Imaging Radiat Oncol. 2025 Mar 8;34:100740. doi: 10.1016/j.phro.2025.100740. eCollection 2025 Apr.
Pancreatic gross tumor volume (GTV) delineation is challenging due to their variable morphology and uncertain ground truth. Previous deep learning-based auto-segmentation methods have struggled to handle tasks with uncertain ground truth and have not accommodated stylistic customizations. We aim to develop a human-in-the-loop pancreatic GTV segmentation tool using Tversky ensembles by leveraging uncertainty estimation techniques.
In this study, we utilized a total of 282 patients from the pancreas task of the Medical Segmentation Decathlon. Thirty patients were randomly selected to form an independent test set, while the remaining 252 patients were divided into an 80-20 % training-validation split. We incorporated Tversky loss layer during training to train a five-member segmentation ensemble with varying contouring tendencies. The Tversky ensemble predicted probability maps by estimating pixel-level segmentation uncertainties. Probability thresholding was employed on the resulting probability maps to generate the final contours, from which eleven contours were extracted for quantitative evaluation against ground truths, with variations in the threshold values.
Our Tversky ensemble achieved DSC of 0.47, HD95 of 12.70 mm and MSD of 3.24 mm respectively using the optimal thresholding configuration. We outperformed the Swin-UNETR configuration that achieved the state-of-the-art result in the pancreas task of the medical segmentation decathlon.
Our study demonstrated the effectiveness of employing an ensemble-based uncertainty estimation technique for pancreatic tumor segmentation. The approach provided clinicians with a consensus probability map that could be fine-tuned in line with their preferences, generating contours with greater confidence.
胰腺大体肿瘤体积（GTV）的勾画具有挑战性，因为其形态多变且真实情况不确定。以往基于深度学习的自动分割方法难以处理真实情况不确定的任务，也无法适应风格定制。我们旨在利用不确定性估计技术，开发一种基于Tversky集成的人工参与的胰腺GTV分割工具。
在本研究中，我们总共使用了医学分割十项全能胰腺任务中的282例患者。随机选择30例患者组成独立测试集，其余252例患者按80%-20%的比例分为训练-验证集。在训练过程中加入Tversky损失层，以训练一个具有不同轮廓倾向的五成员分割集成。Tversky集成通过估计像素级分割不确定性来预测概率图。对生成的概率图采用概率阈值处理以生成最终轮廓，从中提取11个轮廓以根据真实情况进行定量评估，阈值有所不同。
使用最优阈值配置时，我们的Tversky集成分别实现了0.47的DSC、12.70mm的HD95和3.24mm的MSD。我们超过了在医学分割十项全能胰腺任务中取得最先进结果的Swin-UNETR配置。
我们的研究证明了采用基于集成的不确定性估计技术进行胰腺肿瘤分割的有效性。该方法为临床医生提供了一个可根据其偏好进行微调的共识概率图，从而更有信心地生成轮廓。
