Available online 7 April 2023

Author links open overlay panel, , , , , Abstract

High angular resolution diffusion imaging (HARDI) is a promising method for advanced analysis of brain microstructure. However, comprehensive HARDI analysis requires multiple acquisitions of diffusion images (multi-shell HARDI), which is time consuming and often impractical in clinical settings. This study aimed to establish neural network models that can predict new diffusion datasets from clinically feasible brain diffusion MRI for multi-shell HARDI. The development included 2 algorithms: multi-layer perceptron (MLP) and convolutional neural network (CNN). Both followed a voxel-based approach for model training (70%), validation (15%), and testing (15%). The investigations involved 2 multi-shell HARDI datasets: 1) 11 healthy subjects from the Human Connectome Project (HCP); and 2) 10 local subjects with multiple sclerosis (MS). To assess outcomes, we conducted neurite orientation dispersion and density imaging using both predicted and original data and compared their orientation dispersion index (ODI) and neurite density index (NDI) in different brain tissues with 2 measures: peak signal-to-noise ratio (PSNR) and structural similarity index measure (SSIM). Results showed that both models achieved robust predictions, which provided competitive ODI and NDI, especially in brain white matter. The CNN outperformed MLP with the HCP data on both PSNR (p < 0.001) and SSIM (p < 0.01). With the MS data, the models performed similarly. Overall, the optimized neural networks can help generate non-acquired brain diffusion MRI, which will make advanced HARDI analysis possible in clinical practice following further validation. Enabling detailed characterization of brain microstructure will allow enhanced understanding of brain function in both health and disease.

Section snippetsBackground

Diffusion tensor imaging (DTI) is a promising magnetic resonance imaging (MRI) method for characterizing tissue microstructure in both health and disease [1]. However, DTI measures tissue properties on a voxel basis. This makes it limited in resolving complex pathology including inflammatory demyelination and neurodegeneration as seen in neurological diseases such as multiple sclerosis (MS) [2]. Accurate evaluation of tissue pathology is critical for improved disease monitoring and treatment

Data

This study used 2 datasets. The first involved 11 healthy subjects from the WU-Minn cohort of the Human Connectome Project (HCP). The second included 10 people with relapsing-remitting MS (RRMS) recruited locally as part of a multi-site Canadian Prospective Cohort study (CanProCoh). Model development used the 3 T multi-shell HARDI data from HCP. Data acquisition applied a spin-echo echo planar imaging (EPI) sequence, where repetition time (TR) = 5520 ms, echo time (TE) = 89.5 ms, matrix

Model refinement

Both the MLP and CNN models showed a decrease in loss after the early steps of refinement (Fig. 3). The best MLP settings identified during refinement were: 10 non-linear dense layers, 200 neurons per layer, ReLU activation, and zero dropout layers. Accordingly, the best settings for the CNN model were: 2 convolutional/pooling layers, 200 filters per convolutional layer, ReLU activation, average pooling, pool size of 2, and zero batch normalization/dropout layers. Both models predicted

Discussions

Using healthy HCP and clinical MS patient data, this study developed new neural network models for predicting typically unavailable diffusion MRI datasets needed for advanced analysis of brain microstructure in HARDI. Through comprehensive hyperparameter optimization, both the MLP and CNN models showed competitive performance, especially the CNNs, and both improved with training using more subject voxels. In tissue-wise comparisons of brain NODDI excluding the CSF, the CNN-assisted maps were

Conclusion

We demonstrate that robust neural network algorithms can help make advanced HARDI assessment possible using condensed acquisitions of diffusion MRI that are feasible in routine clinical practice. Both the optimized MLP and CNN models show the potential for diffusion MRI prediction as shown using both high-angular HCP and clinical MS data. Based on NODDI analysis, this study also demonstrates the possibility of deriving advanced measures of brain microstructure using predicted data. With further

Funding

This work was supported by the HBI MS Brain and Mental Health Team, University of Calgary, and (in part) by the MS Society of Canada and Canadian Institutes of Health Research. In addition, as part of the Canadian prospective cohort study (CanProCo), the data acquisition from patients was supported by the MS Society of Canada, Brain Canada, Roche, Biogen-Idec, and the Government of Alberta. The healthy scans were obtained from the Human Connectome Project, WU-Minn Consortium (Principal

CRediT authorship contribution statement

Cayden Murray: Methodology, Formal analysis, Data curation, Validation, Writing – original draft, Writing – review & editing. Olayinka Oladosu: Conceptualization, Methodology, Software, Writing – original draft, Writing – review & editing. Manish Joshi: Resources, Supervision, Funding acquisition. Shannon Kolind: Resources, Funding acquisition. Jiwon Oh: Resources, Writing – review & editing, Funding acquisition. Yunyan Zhang: Conceptualization, Methodology, Resources, Writing – review &

Acknowledgements

We thank the patient volunteers for their participation in the study.

References (36)K. Kamiya et al.NODDI in clinical research

J Neurosci Methods

(2020)

M. Rovaris et al.Diffusion MRI in multiple sclerosis

Neurology

(2005)

W.J. Huang et al.Multiple sclerosis: pathology, diagnosis and treatments

Exp Ther Med

(2017)

I. Loma et al.Multiple sclerosis: pathogenesis and treatment

Curr Neuropharmacol

(2011)

M. DescoteauxHigh agular resolution diffusion imaging (HARDI)D. TuchDiffusion MRI of complex tissue structure

(2002)

J.H. Burdette et al.High b-value diffusion-weighted MRI of normal brain

J Comput Assist Tomogr

(2001)

P.B. Kingsley et al.Selection of the optimum b factor for diffusion-weighted magnetic resonance imaging assessment of ischemic stroke

Magn Reson Med

(2004)

View full text

© 2023 Published by Elsevier Inc.