IMPULSED model based cytological feature estimation with U-Net: Application to human brain tumor at 3T

Area of Science:

• Neuroimaging
• Medical Physics
• Artificial Intelligence

Background:

• Diffusion-weighted MRI (DW-MRI) provides insights into tissue microstructure.
• Accurate estimation of microstructural parameters in brain tumors is crucial for diagnosis and treatment planning.
• Current fitting methods can be time-consuming and sensitive to noise.

Purpose of the Study:

• To introduce and validate a deep-learning-based fitting method for rapid and accurate estimation of brain tumor cytological features.
• To utilize the IMPULSED (imaging microstructural parameters using limited spectrally edited diffusion) model with DW-MRI data.
• To compare the deep learning method against conventional non-linear least-squares (NLLS) fitting.

Main Methods:

• A U-Net deep learning model was trained using synthesized image-based data with randomized microstructural parameters.
• The trained U-Net was applied to estimate extracellular diffusion coefficient (Dex), cell size (d), and intracellular volume fraction (vin).
• The method was tested on simulated data and in vivo 3T MRI data from brain tumor patients, comparing performance with NLLS fitting.

Main Results:

• The deep learning method demonstrated superior fidelity and noise robustness compared to NLLS fitting in simulations.
• In vivo data showed improved parameter map quality with the U-Net approach.
• Parameter estimations from U-Net were in good agreement with NLLS fitting.
• The U-Net method achieved a significant speed improvement, reducing computation time from minutes to under a second.

Conclusions:

• The proposed image-based training scheme enhances the quality of estimated microstructural parameters.
• The deep-learning-based fitting method provides fast and accurate estimation of cell microstructural parameters in brain tumors.
• This technique holds potential for improving the efficiency and accuracy of brain tumor analysis using MRI.