基于边界框的 3D AI 模型，用于在标准护理 CT 上对胰腺导管腺癌进行用户引导的容积分割。
Bounding box-based 3D AI model for user-guided volumetric segmentation of pancreatic ductal adenocarcinoma on standard-of-care CTs.
机构信息
Department of Radiology, Mayo Clinic, 200 First Street SW, Rochester, MN, 55905, USA.
Department of Gastroenterology, Hepatology and Nutrition, The University of Texas MD Anderson Cancer Center, 1515 Holcombe Blvd, Houston, TX, 77030, USA.
出版信息
Pancreatology. 2023 Aug;23(5):522-529. doi: 10.1016/j.pan.2023.05.008. Epub 2023 May 26.
OBJECTIVES
To develop a bounding-box-based 3D convolutional neural network (CNN) for user-guided volumetric pancreas ductal adenocarcinoma (PDA) segmentation.
METHODS
Reference segmentations were obtained on CTs (2006-2020) of treatment-naïve PDA. Images were algorithmically cropped using a tumor-centered bounding box for training a 3D nnUNet-based-CNN. Three radiologists independently segmented tumors on test subset, which were combined with reference segmentations using STAPLE to derive composite segmentations. Generalizability was evaluated on Cancer Imaging Archive (TCIA) (n = 41) and Medical Segmentation Decathlon (MSD) (n = 152) datasets.
RESULTS
Total 1151 patients [667 males; age:65.3 ± 10.2 years; T1:34, T2:477, T3:237, T4:403; mean (range) tumor diameter:4.34 (1.1-12.6)-cm] were randomly divided between training/validation (n = 921) and test subsets (n = 230; 75% from other institutions). Model had a high DSC (mean ± SD) against reference segmentations (0.84 ± 0.06), which was comparable to its DSC against composite segmentations (0.84 ± 0.11, p = 0.52). Model-predicted versus reference tumor volumes were comparable (mean ± SD) (29.1 ± 42.2-cc versus 27.1 ± 32.9-cc, p = 0.69, CCC = 0.93). Inter-reader variability was high (mean DSC 0.69 ± 0.16), especially for smaller and isodense tumors. Conversely, model's high performance was comparable between tumor stages, volumes and densities (p > 0.05). Model was resilient to different tumor locations, status of pancreatic/biliary ducts, pancreatic atrophy, CT vendors and slice thicknesses, as well as to the epicenter and dimensions of the bounding-box (p > 0.05). Performance was generalizable on MSD (DSC:0.82 ± 0.06) and TCIA datasets (DSC:0.84 ± 0.08).
CONCLUSION
A computationally efficient bounding box-based AI model developed on a large and diverse dataset shows high accuracy, generalizability, and robustness to clinically encountered variations for user-guided volumetric PDA segmentation including for small and isodense tumors.
CLINICAL RELEVANCE
AI-driven bounding box-based user-guided PDA segmentation offers a discovery tool for image-based multi-omics models for applications such as risk-stratification, treatment response assessment, and prognostication, which are urgently needed to customize treatment strategies to the unique biological profile of each patient's tumor.
目的
开发一种基于边界框的 3D 卷积神经网络（CNN），用于用户引导的胰腺导管腺癌（PDA）的体积分割。
方法
参考分割是在未经治疗的 PDA 的 CT 上（2006-2020 年）获得的。使用以肿瘤为中心的边界框对图像进行算法裁剪，以训练基于 nnUNet 的 3D CNN。三位放射科医生分别在测试子集中对肿瘤进行分割，然后使用 STAPLE 将其与参考分割相结合，得出组合分割。在癌症成像档案（TCIA）（n=41）和医学分割十项全能（MSD）（n=152）数据集上评估了泛化能力。
结果
共有 1151 名患者[667 名男性；年龄：65.3±10.2 岁；T1：34 例，T2：477 例，T3：237 例，T4：403 例；平均（范围）肿瘤直径：4.34（1.1-12.6）-cm]随机分为训练/验证（n=921）和测试子集（n=230；75%来自其他机构）。该模型与参考分割的 DSC（平均值±标准差）较高（0.84±0.06），与复合分割的 DSC（0.84±0.11，p=0.52）相当。模型预测的肿瘤体积与参考体积相似（平均值±标准差）（29.1±42.2-cc 与 27.1±32.9-cc，p=0.69，CCC=0.93）。读者间的变异性较高（平均 DSC 0.69±0.16），尤其是对于较小的等密度肿瘤。相反，该模型在肿瘤分期、体积和密度方面的性能相当（p>0.05）。该模型能够适应不同的肿瘤位置、胰胆管状态、胰腺萎缩、CT 供应商和切片厚度，以及边界框的中心和尺寸（p>0.05）。该模型在 MSD（DSC：0.82±0.06）和 TCIA 数据集（DSC：0.84±0.08）上具有很好的通用性。
结论
基于大型和多样化数据集开发的计算效率高的基于边界框的 AI 模型，对于用户引导的胰腺导管腺癌的体积分割，包括对小的和等密度的肿瘤，表现出了很高的准确性、通用性和对临床遇到的各种变化的鲁棒性。
临床相关性
基于人工智能驱动的边界框的用户引导的 PDA 分割为基于图像的多组学模型提供了一种发现工具，可用于风险分层、治疗反应评估和预后等应用，这些应用迫切需要根据每个患者肿瘤的独特生物学特征来定制治疗策略。
Zotero 插件