Multiclass classification of 18F-DMFP-PET data to assist the diagnosis of parkinsonism

摘要

Parkinson's disease (PD), multiple system atrophy (MSA) and progressive supranuclear palsy (PSP) have similar symptomatology and therefore it is difficult to distinguish among them, especially at early stage. Clinicians normally use different neuroimaging modalities to assist the diagnosis of these disorders, however obtaining an accurate diagnosis is still a challenge. In this work we analyzed a recently emerged neuroimaging modality, 18F-DMFP PET, that allows assessing the deficiency of striatal dopamine that characterizes the parkinsonian syndromes. Three statistical classifiers and several feature extraction approaches were evaluated to automatically differentiate among PD, MSA and PSP. According to the results, PD can be accurately (90% of accuracy) identified using 18F-DMFP PET data, however the identification of MSA and PSP still has room for improvement. Up to our knowledge this is first time that 18F-DMFP PET data is used along with a multiclass classification system for this purpose. In addition, most of the computer systems to assist the diagnosis of parkinsonism only consider the separation of healthy and pathological subjects (binary classification).