Image-based machine learning model as a tool for classification of [ 18 F]PR04.MZ PET images in patients with parkinsonian syndrome

Abstract

Abstract Parkinsonian syndrome (PS) is characterized by bradykinesia, resting tremor, and rigidity, and it represents the phenotype observed in various neurodegenerative disorders. Positron emission tomography (PET) imaging plays an important role in diagnosing PS by detecting the progressive loss of dopaminergic neurons. This study aimed to develop and compare five machine-learning models for classifying [18F]PR04.MZ PET images between patients with PS and subjects without evidence for dopaminergic deficit (SWEDD). A dataset of [18F]PR04.MZ PET images from 204 subjects was analyzed and classified into PS compatible (1) and SWEDDs (0) by three blinded expert readers. The images were preprocessed to generate two and three-dimensional datasets. Five different pattern recognition algorithms, commonly used for image analysis, were trained and validated, comparing their performance to the majority reading of expert diagnosis considered as the standard of truth. Three models outperformed the others, achieving an accuracy greater than 98%. The results demonstrated that our machine-learning models, combined with [18F]PR04.MZ PET images, provide highly accurate and precise tools to support clinicians in PET image analysis. This approach may reduce the time required for interpretation and increase certainty in the diagnostic process.