Noninvasive molecular subtyping of pediatric low-grade glioma with self-supervised transfer learning.

Researchers

Journal

Modalities

Models

Abstract

To develop and externally validate a scan-to-prediction deep-learning pipeline for noninvasive, MRI-based BRAF mutational status classification for pLGG.We conducted a retrospective study of two pLGG datasets with linked genomic and diagnostic T2-weighted MRI of patients: Boston Children’s Hospital (development dataset, N=214), and Child Brain Tumor Network (CBTN) (external validation, N=112). We developed a deep learning pipeline to classify BRAF mutational status (V600E vs. fusion vs. wild-type) from whole-scan input via a two-stage process: 1) 3D tumor segmentation and extraction of axial tumor images, and 2) slice-wise, deep learning-based classification of mutational status. We investigated knowledge-transfer approaches to prevent model overfitting with a primary endpoint of the area under the receiver operating characteristic curve (AUC). To enhance model interpretability, we developed a novel metric, COMDist that quantifies the accuracy of model attention with respect to the tumor.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 AUC, taken as a weighted average across the three mutational classes, (0.82 [95% CI: 0.70-0.90], Accuracy: 77%) on internal validation and (0.73 [95% CI 0.68-0.88], Accuracy: 75%) on external validation. Training with TransferX also led to an AUC improvement of 17.7% and a COMDist Improvement of 6.42% over training from scratch on the development dataset.Transfer learning and self-supervised cross-training improved classification performance and generalizability for noninvasive pLGG mutational status prediction in a limited data scenario.The authors developed and externally validated an automated, scan-to-prediction deep learning pipeline that classifies BRAF Mutational status in pediatric low-grade gliomas directly from T2-Weighted MRI scans.An innovative training approach combining self-supervision and transfer learning (“TransferX”) is developed to boost model performance in low data settings;TransferX enables the development of a scan-to-prediction pipeline for pediatric LGG mutational status (BRAF V600E, fusion, or wildtype) with high accuracy and mild performance degradation on external validation;An evaluation metric “COMDist” is proposed to increase interpretability and quantify the accuracy of the model’s attention around the tumor.