基于自监督迁移学习的小儿低级别胶质瘤无创分子分型。

Noninvasive Molecular Subtyping of Pediatric Low-Grade Glioma with Self-Supervised Transfer Learning.

机构信息

From the Artificial Intelligence in Medicine Program, Mass General Brigham, Harvard Medical School, Boston, Mass (D.T., Z.Y., A.Z., Y.Z., A.B., R.C., H.H., H.J.W.L.A., B.H.K.); Department of Radiation Oncology, Dana-Farber Cancer Institute, Brigham and Women's Hospital, Boston Children's Hospital, Harvard Medical School, 75 Francis St, Boston, MA 02115 (D.T., Z.Y., A.Z., Y.Z., A.B., R.C., H.H., K.X.L., H.E., H.J.W.L.A., D.A.H.K., B.H.K.); Department of Radiology, Boston Children's Hospital, Harvard Medical School, Boston, Mass (S.V., S.P.P., T.Y.P.); Center for Data-Driven Discovery in Biomedicine (A.N., A.F.) and Department of Neurosurgery (A.F., A.C.R.), Children's Hospital of Philadelphia, Philadelphia, Pa; Department of Radiology, Perelman School of Medicine, University of Pennsylvania, Philadelphia, Pa (A.N.); Departments of Neurology, Pediatrics, and Neurologic Surgery, University of California San Francisco, San Francisco, Calif (S.M.); Department of Radiology, Brigham and Women's Hospital, Dana-Farber Cancer Institute, Harvard Medical School, Boston, Mass (H.J.W.L.A.); Department of Radiology and Nuclear Medicine, CalifRIM & GROW, Maastricht University, Maastricht, the Netherlands (H.J.W.L.A.); and Department of Pediatric Oncology (P.B.) and Department of Pathology (K.L.L.), Dana-Farber Cancer Institute, Boston Children's Hospital, Harvard Medical School, Boston, Mass.

出版信息

Radiol Artif Intell. 2024 May;6(3):e230333. doi: 10.1148/ryai.230333.

DOI:10.1148/ryai.230333

PMID:38446044

原文链接:https://pmc.ncbi.nlm.nih.gov/articles/PMC11140508/

Abstract

Purpose To develop and externally test a scan-to-prediction deep learning pipeline for noninvasive, MRI-based mutational status classification for pediatric low-grade glioma. Materials and Methods This retrospective study included two pediatric low-grade glioma datasets with linked genomic and diagnostic T2-weighted MRI data of patients: Dana-Farber/Boston Children's Hospital (development dataset, = 214 [113 (52.8%) male; 104 (48.6%) wild type, 60 (28.0%) fusion, and 50 (23.4%) V600E]) and the Children's Brain Tumor Network (external testing, = 112 [55 (49.1%) male; 35 (31.2%) wild type, 60 (53.6%) fusion, and 17 (15.2%) V600E]). A deep learning pipeline was developed to classify mutational status ( wild type vs fusion vs V600E) via a two-stage process: three-dimensional tumor segmentation and extraction of axial tumor images and section-wise, deep learning-based classification of mutational status. Knowledge-transfer and self-supervised approaches were investigated to prevent model overfitting, with a primary end point of the area under the receiver operating characteristic curve (AUC). To enhance model interpretability, a novel metric, center of mass distance, was developed to quantify the model attention around the tumor. Results A combination of transfer learning from a pretrained medical imaging-specific network and self-supervised label cross-training (TransferX) coupled with consensus logic yielded the highest classification performance with an AUC of 0.82 (95% CI: 0.72, 0.91), 0.87 (95% CI: 0.61, 0.97), and 0.85 (95% CI: 0.66, 0.95) for wild type, fusion, and V600E, respectively, on internal testing. On external testing, the pipeline yielded an AUC of 0.72 (95% CI: 0.64, 0.86), 0.78 (95% CI: 0.61, 0.89), and 0.72 (95% CI: 0.64, 0.88) for wild type, fusion, and V600E, respectively. Conclusion Transfer learning and self-supervised cross-training improved classification performance and generalizability for noninvasive pediatric low-grade glioma mutational status prediction in a limited data scenario. Pediatrics, MRI, CNS, Brain/Brain Stem, Oncology, Feature Detection, Diagnosis, Supervised Learning, Transfer Learning, Convolutional Neural Network (CNN) © RSNA, 2024.

摘要

目的开发并外部测试一种基于深度学习的扫描到预测的管道，用于对儿科低级别胶质瘤进行非侵入性、基于 MRI 的突变状态分类。

材料与方法本回顾性研究纳入了两个具有患者基因组和诊断性 T2 加权 MRI 数据链接的儿科低级别胶质瘤数据集：Dana-Farber/Boston 儿童医院（开发数据集，=214 [113 例（52.8%）为男性；104 例（48.6%）野生型，60 例（28.0%）融合型，50 例（23.4%）V600E]）和儿童脑肿瘤网络（外部测试，=112 [55 例（49.1%）为男性；35 例（31.2%）野生型，60 例（53.6%）融合型，17 例（15.2%）V600E]）。开发了一种深度学习管道，通过两阶段过程对突变状态（野生型与融合型与 V600E）进行分类：三维肿瘤分割和提取轴向肿瘤图像，以及基于节段的突变状态深度学习分类。研究了知识转移和自我监督方法以防止模型过拟合，主要终点为接收器工作特征曲线（AUC）下的面积。为了增强模型的可解释性，开发了一种新的度量标准，即质心距离，用于量化模型在肿瘤周围的注意力。

结果结合来自预训练的医学成像特定网络的迁移学习和自我监督标签交叉训练（TransferX），并结合共识逻辑，在内部测试中，分类性能最高，野生型的 AUC 为 0.82（95%CI：0.72，0.91），融合型为 0.87（95%CI：0.61，0.97），V600E 为 0.85（95%CI：0.66，0.95）。在外部测试中，该管道在野生型、融合型和 V600E 中的 AUC 分别为 0.72（95%CI：0.64，0.86）、0.78（95%CI：0.61，0.89）和 0.72（95%CI：0.64，0.88）。

结论在有限的数据情况下，迁移学习和自我监督的交叉训练提高了非侵入性儿科低级别胶质瘤突变状态预测的分类性能和泛化能力。

儿科；磁共振成像；中枢神经系统；脑/脑干；肿瘤学；特征检测；诊断；有监督学习；迁移学习；卷积神经网络（CNN）