Presentation Information
[SS24-05]Deep Learning Based Super-Resolution of Diffusion MRI with Applications to Prostate Cancer and TBI
*Hirak Doshi1, Siddharth Singh2, Supriya Devi Phurailatpam2, Sudhir Kumar Pathak3, Durgesh K Dwivedi2, Rathish Kumar B V1 (1. Department of Mathematics and Statistics, IIT Kanpur (India), 2. Department of Radiodiagnosis, King George’s Medical University, Lucknow (India), 3. Learning Research and Development Centre, University of Pittsburgh (United States of America))
Keywords:
deep learning,diffusion mri,super resolution
Prostate cancer is one of the most common malignancies affecting men globally, originating in the prostate gland where it often manifests as a slow-growing tumor, though aggressive forms can progress rapidly. Recent advances in magnetic resonance imaging have positioned multi-shell diffusion MRI (dMRI) as a promising tool for improving prostate cancer diagnosis and characterization. By acquiring data over multiple b-values, multi-shell dMRI enables the capture of diverse diffusion environments within the prostate, facilitating a detailed assessment of tissue microstructure. Contemporary techniques now use multi-compartment models such as Restriction Spectrum Imaging (RSI) to better understand the contributions of restricted, hindered and free diffusion components with recent studies indicating improved lesion conspicuity, sensitivity and specificity compared with conventional apparent diffusion coefficient (ADC) mapping.
Super-resolution in multi-shell diffusion MRI offers a transformative approach to overcoming the intrinsic trade-off between spatial resolution and signal-to-noise ratio. Acquiring high-resolution diffusion images is often challenging due to increased scan times, patient motion and reduced SNR at high b-values. However, advanced super-resolution techniques allow reconstruction of images with sub-voxel accuracy from lower-resolution acquisitions.
Our method synthesizes information across multiple diffusion shells reflecting distinct microstructural properties to generate high-resolution images that reveal subtle tissue details, thereby providing more accurate quantitative metrics for distinguishing malignant from benign tissue. Similar advanced imaging techniques are also being explored in the context of traumatic brain injury (TBI) where they show promise for detecting microstructural alterations following injury highlighting a broader applicability of this approach.
Super-resolution in multi-shell diffusion MRI offers a transformative approach to overcoming the intrinsic trade-off between spatial resolution and signal-to-noise ratio. Acquiring high-resolution diffusion images is often challenging due to increased scan times, patient motion and reduced SNR at high b-values. However, advanced super-resolution techniques allow reconstruction of images with sub-voxel accuracy from lower-resolution acquisitions.
Our method synthesizes information across multiple diffusion shells reflecting distinct microstructural properties to generate high-resolution images that reveal subtle tissue details, thereby providing more accurate quantitative metrics for distinguishing malignant from benign tissue. Similar advanced imaging techniques are also being explored in the context of traumatic brain injury (TBI) where they show promise for detecting microstructural alterations following injury highlighting a broader applicability of this approach.