Deep learning-based prediction of diffusion MRI responses from microvascular OCT angiograms.
Authors
Affiliations (5)
Affiliations (5)
- Department of Computer Science & Software Engineering, United Arab Emirates University, Sheik Khalifa Bin Zayed Street, Asharej, Al Ain, 15551, United Arab Emirates.
- Department of Computer Science & Software Engineering, United Arab Emirates University, Sheik Khalifa Bin Zayed Street, Asharej, Al Ain, United Arab Emirates.
- Biomedical Engineering Institute, École Polytechnique de Montréal, Montreal, H3T 1J4, Montreal, Quebec, H3T 1J4, Canada.
- Department of Computer Science and Software Engineering, United Arab Emirates University, Sheik Khalifa Bin Zayed Street, Asharej, Al Ain, Abu Dhabi, 15551, United Arab Emirates.
- Department of Computer Science and Software Engineering, United Arab Emirates University, Sheik Khalifa Bin Zayed Street, Asharej, Al Ain, 15551, United Arab Emirates.
Abstract
<i>Objective.</i>Diffusion magnetic resonance imaging (MRI) signals are sensitive to microvascular geometry and hemodynamics, but current approaches rely on computationally intensive simulations on optical coherence tomography (OCT)–derived vascular graphs. We investigated whether three‐dimensional deep learning models can serve as fast surrogates that predict diffusion MRI responses directly from microvascular OCT angiograms.

<i>Approach.</i>Five mouse cortical OCT angiograms were segmented into 4,792 vascular subgraphs and converted into nine‐channel image volumes encoding vascular geometry, magnetic field gradient, hematocrit, oxygen saturation, partial oxygen pressure, blood velocity magnitude, and three velocity components. For each sample, a physics‐based framework generated spin echo (SE) and diffusion‐weighted signals for three<i>b</i>-values (50, 100, 500 s/mm<sup>2</sup>), three gradient timing pairs, and ten gradient directions. We trained five 3D architectures (AutoEncoder, BasicUNet, DenseNet169, EfficientNetB4 and ResNet) with late fusion of acquisition parameters and evaluated performance using MSE, MAE and<i>R</i><sup>2</sup>.

<i>Main results.</i>Under within‐distribution (random‐split) evaluation, test<i>R</i><sup>2</sup>values were at least 0.87. DenseNet169 achieved the best overall accuracy (MSE = 0.0039 ± 0.0029, MAE = 0.0397 ± 0.0161,<i>R</i><sup>2</sup>= 0.92) and maintained<i>R</i><sup>2</sup>> 0.80 across all diffusion settings, while ResNet provided near‐perfect reconstruction of the SE sequence alone (<i>R</i><sup>2</sup>≈ 1.00). Leave‐one‐subject‐out cross‐validation gave MSE = 0.0173 ± 0.0012, MAE = 0.094 ± 0.004, and<i>R</i><sup>2</sup>= 0.666 ± 0.019, with 34.4% of held‐out samples reaching<i>R</i><sup>2</sup>≥ 0.80. Channel ablation revealed velocity magnitude and velocity components as dominant predictors. Once trained, the models predicted full diffusion responses in milliseconds, providing speedups of four to six orders of magnitude compared with simulations.

<i>Significance.</i>These results demonstrate that deep learning surrogates can accurately and efficiently approximate diffusion MRI responses from microvascular OCT angiograms, enabling rapid exploration of microvascular configurations and acquisition protocols.