Toward Noninvasive High-Resolution In Vivo pH Mapping in Brain Tumors by ³¹P-Informed deepCEST MRI

Area of Science:

• Medical Imaging
• Biophysics
• Artificial Intelligence

Background:

• Intracellular pH (pH_i) is crucial for understanding brain tumor pathologies.
• Conventional ³¹P-MRS for pH_i measurement has limitations in spatial resolution and scan time.
• ¹H-based APT-CEST imaging offers improved resolution and faster scanning for pH_i estimation.

Purpose of the Study:

• To develop and validate a fully connected neural network for direct prediction of ³¹P-pH_i maps from ¹H-based CEST data.
• To compare the performance of the novel deep learning approach against conventional CEST metrics and ³¹P-MRS.
• To assess the capability of the model in revealing tumor heterogeneity and capturing unique pH information.

Main Methods:

• A fully connected neural network was trained voxel-wise using CEST and T1 data from 11 brain tumor patients.
• The network predicted ³¹P-pH_i values, with validation performed on 4 test patients.
• Predicted pH_i maps were compared with ³¹P-MRS measurements and conventional CEST metrics.

Main Results:

• The deep learning model demonstrated a general correspondence between predicted and measured ³¹P-pH_i maps, with a RMSE of 0.04 pH units in tumor regions.
• High-resolution predictions revealed tumor heterogeneity not apparent in conventional CEST data.
• The model successfully leveraged APT-CEST's hidden pH-sensitivity, providing higher resolution pH_i maps than ³¹P-MRS.

Conclusions:

• The developed deepCEST pH_i neural network enables higher spatial resolution pH_i mapping with shorter scan times compared to ³¹P-MRS.
• This approach offers a promising tool for 3D pH imaging in clinical settings.
• Future studies can enhance the model by incorporating additional CEST features for improved accuracy and clinical utility.