Research Article

Classification of Parkinsonian Syndromes from FDG-PET Brain Data Using Decision Trees with SSM/PCA Features

Table 4

Performance for binary classification of disease groups (number of cases in brackets) in the LOOCV with feature preselection. The columns Feat. and Perf. indicate the percentage of features used and the corresponding performance. The remaining columns show confusion matrices and class accuracies. The number of subjects correctly classified for each class is in bold.

Feat. % Perf. % Class PD (20) PSP (17) MSA (21)

100 65.5PD 1553
PSP 483
MSA 1415
Accuracy7547.171.4

50 67.2PD 1552
PSP 494
MSA 1315
Accuracy7552.971.4

25 69PD 1552
PSP 493
MSA 1316
Accuracy7552.976.2