Prediction of 123I-FP-CIT SPECT Results from First Acquired Projections Using Artificial Intelligence
Wadi' Othmani1, Arthur Coste1, Dimitri Papathanassiou1,2,3
1Médecine Nucléaire, Institut Godinot, 51100 Reims, France.
Diagnostics (Basel, Switzerland)
|June 13, 2025
Summary
A new Convolutional Neural Network (CNN) can predict Parkinsonian syndrome diagnosis from early dopamine transporter imaging, aiding atypical cases. This AI tool reliably detects normal patients, improving diagnostic efficiency.
Area of Science:
- Nuclear medicine
- Artificial intelligence in medical imaging
Background:
- 123I-FP-CIT SPECT imaging is crucial for diagnosing Parkinsonian syndromes, especially in atypical presentations.
- Maintaining patient immobility during SPECT acquisition is challenging for some patients.
Purpose of the Study:
- To develop a Convolutional Neural Network (CNN) to predict SPECT scan outcomes from initial projections.
- To enable reliable detection of normal patients using AI.
Main Methods:
- A VGG16-like CNN was trained on 982 123I-FP-CIT SPECT scans.
- The model predicted abnormality probability from the first projection, incorporating patient age.
- Model performance was validated on an independent test set of 100 scans.
Main Results:
- The CNN achieved 98.0% sensitivity and 96.3% negative predictive value.
- Overall accuracy was 75.0%, with saliency maps highlighting basal ganglia.
- The model demonstrated high reliability in excluding presynaptic dopaminergic loss.
Conclusions:
- A trained CNN can reliably predict presynaptic dopaminergic loss from early SPECT projections.
- This AI approach may benefit patients with limited compliance.
- Further validation across multiple centers is recommended.
