An Automatic Striatum Segmentation Model to Estimate MR Maps for Dopamine Transporter SPECT using Deep Learning

Motivation: Striatum segmentation on SPECT is necessary to quantify uptake for Parkinson's disease (PD), but is challenging due to the inferior resolution. MRI is the preferred reference for segmentation due to its excellent soft tissue contrast. Goal(s): This work proposes cross-modality automatic striatum segmentation, estimating MR striatal maps from clinical SPECT using deep learning (DL). Approach: nnU-Net-based method are implemented and SPECT images are paired with MR-based striatal maps as supervised learning (training:validation:testing = 136:24:40)Results: The proposed method can segment 4 MR-like individual compartments on clinical SPECT, which is also superior to several traditional and DL methods, both in physical and clinical metrics. Impact: The proposed DL-based cross-modality striatum segmentation method is feasible for clinical DaT SPECT in PD, and 4 MR-like individual compartments can be obtained to quantify striatal uptake, which is beneficial to the accurate diagnosis and clinical management of PD.