Accurate Automated Quantification of Dopamine Transporter PET Without MRI Using Deep Learning-based Spatial...

Area of Science:

• Neurology
• Medical Imaging
• Artificial Intelligence

Background:

• Dopamine transporter imaging is vital for diagnosing Parkinson's disease (PD) and related disorders.
• ¹⁸F-FP-CIT PET offers superior resolution and sensitivity compared to SPECT.
• Accurate quantification of ¹⁸F-FP-CIT PET is crucial for clinical assessment.

Purpose of the Study:

• To develop a novel, automatic quantification method for ¹⁸F-FP-CIT PET images.
• To utilize artificial intelligence (AI)-based spatial normalization (SN) technology that eliminates the need for anatomical images.
• To enable accurate assessment of presynaptic dopaminergic function.

Main Methods:

• Developed an AI-based SN engine using convolutional neural networks trained on 213 paired ¹⁸F-FP-CIT PET and 3D MRI datasets.
• Employed a cyclic training strategy for backward deformation from template to individual space.
• Validated the method's accuracy using 89 internal and 135 external paired ¹⁸F-FP-CIT PET and MRI datasets, comparing with MRI-based quantification (FIRST software).

Main Results:

• The AI-based method successfully generated spatially normalized ¹⁸F-FP-CIT PET images without requiring CT or MRI.
• High correlation (R² 0.96-0.99, slope 0.98-1.02) was observed between the PET-only method and MRI-based quantification across internal and external datasets.
• Demonstrated accurate quantification of striatal activity.

Conclusions:

• The AI-based SN method provides accurate, automatic quantification of striatal activity in ¹⁸F-FP-CIT brain PET images.
• This approach eliminates the need for MRI support, simplifying the imaging process.
• The method shows significant promise for evaluating presynaptic dopaminergic function in PD and related parkinsonian disorders.