Detecting IDH and TERTp mutations in diffuse gliomas using ¹H-MRS with attention deep-shallow networks

Area of Science:

• Neuro-oncology
• Medical imaging
• Computational biology

Background:

• Accurate preoperative identification of isocitrate dehydrogenase (IDH) and telomerase reverse transcriptase gene promoter (TERTp) mutations in glioma is crucial for patient prognosis and treatment strategies.
• Current methods may be invasive or lack precision, highlighting the need for advanced noninvasive diagnostic tools.

Purpose of the Study:

• To develop and evaluate deep learning classifiers for noninvasively identifying IDH and TERTp mutations in hemispheric diffuse gliomas.
• To utilize proton magnetic resonance spectroscopy (1H-MRS) data with a one-dimensional convolutional neural network (1D-CNN) architecture for mutation detection.

Main Methods:

• Analysis of 1H-MRS data from 225 adult patients with hemispheric diffuse glioma.
• Processing of spectral data using LCModel and training of various deep learning models, including an Attention Deep-Shallow Network (ADSN).
• Application of Gradient-weighted Class Activation Mapping (Grad-CAM) for model interpretability.

Main Results:

• The ADSN model demonstrated high accuracy in IDH mutation detection, achieving F1-scores of 93% (validation) and 88% (test).
• For TERTp mutation detection, the ADSN model achieved F1-scores of 80% (validation) and 81% (test).
• ADSN achieved 88% (validation) and 86% (test) F1-scores for detecting TERTp-only gliomas.

Conclusions:

• Deep learning models, particularly ADSN, can accurately predict IDH and TERTp mutational status in diffuse gliomas using 1H-MRS data.
• The models effectively extract relevant information from spectra, eliminating the need for manual feature engineering.
• This noninvasive deep learning approach shows significant potential for improving glioma diagnosis and management.