Parametric cerebral blood flow and arterial transit time mapping using a 3D convolutional neural network

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Neuroscience

Background:

• Multi-post-labeling delay (multi-PLD) pseudo-continuous arterial spin labeling (pCASL) is crucial for assessing cerebral blood flow (CBF) and arterial transit time (ATT).
• Conventional methods require long scan times due to multiple PLDs and averages, limiting clinical applicability.
• Reducing scan time without compromising accuracy is a key challenge in pCASL.

Purpose of the Study:

• To develop a hierarchically structured 3D convolutional neural network (H-CNN) to shorten the scan time of multi-PLD pCASL.
• To enable accurate estimation of ATT and CBF maps using a reduced number of PLDs and averages.
• To evaluate the performance of the H-CNN against conventional methods.

Main Methods:

• A total of 48 subjects underwent multi-PLD pCASL MRI.
• A novel H-CNN was developed to estimate ATT and CBF maps from reduced PLDs and averages.
• The H-CNN's performance was compared to a nonlinear model fitting method using mean absolute error (MAE).

Main Results:

• The H-CNN achieved low MAEs for ATT (32.69 ms) and CBF (3.32 mL/100 g/min) with a full dataset (six PLDs, six averages).
• Even with a reduced dataset (three PLDs), the H-CNN estimated ATT and CBF with minimal discrepancy from the reference values (MAEs of 231.45 ms for ATT and 9.80 mL/100 g/min for CBF).

Conclusions:

• The proposed machine learning-based approach using H-CNN substantially reduces the scan time for multi-PLD pCASL.
• This method offers a promising solution for efficient and accurate ATT and CBF mapping in clinical settings.
• The H-CNN facilitates faster acquisition of critical hemodynamic information from brain imaging.