Deep Learning-Based Inversion for Diffusion-Relaxation Correlation Spectrum and its Application in Prostate Cancer Diagnosis

Motivation: Diffusion-relaxation correlation spectrum imaging (DR-CSI) allows for sub-voxel study of prostate microstructures. However, traditional inversion methods for DR-CSI are both unstable and computationally intensive. Goal(s): To improve the speed and robustness of DR-CSI inversion with deep learning, and apply it to prostate cancer (PCa) detection. Approach: Models for image denoising and inversion were trained with synthesized DR-CSI data. Results: The peak distribution regions of the T2-ADC spectra within prostate cancer (PCa) regions exhibit substantial differences compared to those in non-PCa regions. Component analysis of T2-ADC spectra has potential in the diagnosis of prostate cancer. Impact: Deep learning can accelerates the inversion substantially and increase the robustness of results, which can facilitate the application of sub-voxel analysis based on DR-CSI.