Logistic discriminant parametric mapping: a novel method for the pixel-based differential diagnosis of Parkinson's disease

Positron emission tomography (PET) and single-photon emission tomography (S PET) imaging of the dopaminergic system is a powerful tool for distinguishi ng groups of patients with neurodegenerative disorders, such as Parkinson's disease (PD). However, the differential diagnosis of individual subjects p resenting early in the progress of the disease is much more difficult, part icularly using region-of-interest analysis where small localized difference s between subjects are diluted. In this paper we present a novel pixel-base d technique using logistic discriminant analysis to distinguish between a g roup of PD patients and age-matched healthy controls. Simulated images of a n anthropomorphic head phantom were used to test the sensitivity of the tec hnique to striatal lesions of known size. The methodology was applied to re al clinical SPET images of binding of technetium-99m labelled TRODAT-1 to d opamine transporters in PD patients (n=42) and age-matched controls (n=23). The discriminant model was trained on a subset (n=17) of patients for whom the diagnosis was unequivocal. Logistic discriminant parametric maps were obtained for all subjects, showing the probability distribution of pixels c lassified as being consistent with PD. The probability maps were corrected for correlated multiple comparisons assuming an isotropic Gaussian point sp read function. Simulated lesion sizes measured by logistic discriminant par ametric mapping (LDPM) gave strong correlations with the known data (r(2)=0 .985, P<0.001). LDPM correctly classified all PD patients (sensitivity 100% ) and only misclassified one control (specificity 95%). All patients who ha d equivocal clinical symptoms associated with early onset PD (n=4) were cor rectly assigned to the patient group. Statistical parametric mapping (SPM) had a sensitivity of only 24% on the same patient group. LDPM is a powerful pixel-based tool for the differential diagnosis of patients with PD and he althy controls. The diagnosis of disease even before clinical symptoms beco me apparent may be possible, and ultimately this technique could be most us eful in differentiating between several neurodegenerative disorders, incorp orating images of multiple neuroreceptor systems.