An Accelerated Spectroscopic MRI Metabolite Quantification Based on a Deep Learning Method for Radiation Therapy...

Area of Science:

• Neuroimaging
• Quantitative MRI
• Artificial Intelligence in Medicine

Background:

• Spectroscopic MRI (sMRI) quantifies brain tumors without contrast agents.
• Previous studies show sMRI-guided radiation therapy improves patient survival.
• Current spectral fitting methods (FITT) are computationally intensive.

Purpose of the Study:

• To compare the performance of NNFit, a deep learning model, against FITT for sMRI spectral analysis.
• To evaluate NNFit's accuracy in metabolite quantification and its impact on radiation treatment planning.

Main Methods:

• NNFit, a self-supervised deep learning model, was trained on sMRI data to estimate choline (Cho), creatine (Cr), and NAA levels.
• The model was trained on 30 GBM patients (56 scans) and tested on 17 GBM patients (29 scans).
• Performance was evaluated against FITT using structural similarity indices (SSIM) and Dice coefficient.

Main Results:

• NNFit demonstrated significantly faster processing speeds compared to FITT.
• Metabolite quantification showed strong agreement between NNFit and FITT.
• Radiation target volumes derived from NNFit were visually comparable and had fewer artifacts.

Conclusions:

• NNFit offers a rapid, accurate, and artifact-reduced approach for metabolic imaging.
• The model shows significant potential for accelerating radiotherapy planning in clinical settings.