Generation of 123I-IMP brain SPECT from 3D T1-weighted imaging using a machine-learning-based model

Motivation: 123I-IMP SPECT is useful for diagnosing dementia and Parkinsonism but has drawbacks. We explored whether MPRAGE images could generate SPECT-like images, reducing the need for traditional SPECT. Goal(s): This study aimed to develop an ML model that generates SPECT-like images from MPRAGE, using a dataset of dementia and Parkinsonism patients to improve model accuracy. Approach: We used the Pix2PixHD method with various smoothing levels on MPRAGE images and compared models trained on dementia, Parkinsonism, and combined datasets. Quantitative and qualitative assessments were performed. Results: The combined dataset achieved the highest accuracy, showing that training on diverse cases improves image generation and model generalizability. Impact: This study shows that a machine learning model trained on a combined dementia and Parkinsonism dataset can generate accurate SPECT-like images from MPRAGE, potentially reducing the need for traditional SPECT imaging in neurological assessments.