Machine learning, radiomics distinguishes glioma

After developing a workflow to support this, researchers at the Yale School of Medicine created an automated approach that segments gliomas on MR exams of the brain, performs radiomics analysis, and then predicts whether the tumor is high or low grade. When tested, their approach yielded an area under the curve (AUC) of 0.86.

“We were able to develop a PACS-based auto-segmentation tool, which coupled with a high- vs. low-grade glioma prediction tool,” said Sara Merkaj, a postdoctoral researcher. “This algorithm could potentially be incorporated into clinical practice.”

Gliomas are the most common primary brain tumors and are classified – based on pathology and molecular markers – as grade 1-4 according to World Health Organization (WHO) criteria. Grades 1-2 are considered low grade gliomas while grades 3-4 are considered high grades.

“There is a difference in prognosis and treatment between high- and low-grade gliomas, and the current gold standard for assessment is biopsy, which is long-lasting and invasive and involves certain surgical needs,” she said. “For that reason, there is a great need for non-invasive, preoperative glioma prediction tools.”

To develop their model, the researchers used the dataset from the Brain Tumor Segmentation (BraTS) 2021 Challenge to pre-train a U-Net algorithm to perform auto-segmentation of whole, nuclear, and necrotic segments of primary brain tumors on fluid attenuated inversion repair (FLAIR) MR images. They then validated the model using pre-therapy MRI scans of patients from Yale New Haven Health.

After nearly 2,000 radiomics features were extracted using the PyRadiomics software from three regions of interest and six sequences, an XGBoost machine learning algorithm was then trained to classify high- vs. low-grade gliomas. The researchers internally validated the model with 30 replicates of five-fold cross-validation.

In addition, the researchers used a UNet Transformer deep-learning architecture to embed their algorithm into PACS software (Visage 7, Visage Imaging† The PyRadiomics software was also embedded in Visage 7 to extract radiomic functions from the image segmentations.

When tested with 324 high- and low-grade gliomas, the algorithm yielded a mean AUC of 0.86.

In future directions, the researchers are currently exploring new methods for dealing with missing data and unbalanced data sets, Merkaj said. In addition, they plan to test this algorithm in a prospective clinical trial, with the aim of incorporating it into clinical practice later, she said.

In addition, they are currently investigating the use of deep-learning algorithms for glioma prediction. In addition, they want to combine predictions of tumor grade with molecular subtypes, especially isocitrate dehydrogenase (IDH) status, Merkaj said.

Copyright © 2022

Leave a Comment

Your email address will not be published. Required fields are marked *