Exploring the Potential of Machine Learning Algorithms to Improve Diffusion Nuclear Magnetic Resonance Imaging Models...

Area of Science:

• Neuroimaging
• Biomedical Engineering
• Computational Biology

Background:

• Diffusion nuclear magnetic resonance imaging (dMRI) is crucial for understanding tissue microstructure.
• Analytical fitting methods for dMRI models can face limitations, especially with complex diffusion patterns.
• Machine learning (ML) presents a promising avenue for overcoming these analytical challenges.

Purpose of the Study:

• To explore and evaluate various ML algorithms for dMRI model analysis.
• To compare the performance of ML techniques against conventional analytical methods.
• To assess the efficiency and accuracy of ML in predicting and estimating dMRI parameters.

Main Methods:

• Trained and tested ML classification and regression algorithms on dMRI datasets.
• Utilized algorithms such as Extra-Tree Classifier, Multilayer Perceptron, and Random Forest.
• Evaluated performance using accuracy, AUC, RMSE_CV, and computational timing.

Main Results:

• Extra-Tree Classifier and Multilayer Perceptron achieved high accuracy (94.1% and 91.7%) and AUC (98.7% and 96.3%) for classification.
• Random Forest provided the most accurate parameter estimation with low RMSE_CV percentages for D, D*, f, and K.
• ML methods were approximately 232 times faster than conventional methods post-training.

Conclusions:

• ML algorithms significantly enhance the efficiency and accuracy of dMRI analysis.
• ML offers novel perspectives on the microstructural and functional organization of biological tissues.
• The study highlights the potential of ML for advancing dMRI research and applications.