Towards dopaminergic SPECT estimation from structural MRI sequences to characterize Parkinson's disease

Area of Science:

• Neuroimaging
• Artificial Intelligence
• Biomedical Engineering

Background:

• Parkinson's disease involves progressive dopaminergic neuron loss, leading to motor deficits.
• Current diagnosis often occurs late, after significant neuronal damage (80% loss).
• Existing neuroimaging methods like T1-MRI reveal structural changes, while SPECT sequences can measure dopamine activity but are limited in availability and subjective analysis.

Purpose of the Study:

• To develop a self-supervised deep learning model for estimating dopaminergic levels from standard MRI scans.
• To create an MRI-to-SPECT transformation for capturing dopamine-related information.
• To enable classification between Parkinson's disease patients and healthy controls using learned representations.

Main Methods:

• A self-supervised deep representation learning approach was employed.
• The model learned an embedding representation through an MRI-to-SPECT translation task.
• A retrospective dataset of 73 Parkinson's patients and 73 controls was utilized for training and validation.

Main Results:

• The proposed method achieved a precision of 79% in discriminating between Parkinson's and control subjects.
• The self-supervised representation effectively captured dopaminergic level information from MRI data.
• Demonstrated the potential of transforming MRI data into a proxy for dopamine activity.

Conclusions:

• A novel self-supervised deep learning framework can estimate dopaminergic levels from routine MRI scans.
• This approach offers a potential non-invasive method for early Parkinson's disease detection and diagnosis.
• The MRI-to-SPECT translation task provides a valuable tool for analyzing dopamine function in neurodegenerative diseases.