J Nucl Med. 2008; 49 (Supplement 1):377P
Instrumentation & Data Analysis: Data Analysis & ManagementData Analysis & Management Posters |
A multiclass probabilistic diagnosis of dementia using data mining techniques
Michael Bewley1,
Lingfeng Wen1,
Stefan Eberl2,
Michael Fulham2 and
Dagan Feng3
1 University of Sydney, Sydney, New South Wales, Australia;
2 Royal Prince Alfred Hospital, Sydney, New South Wales, Australia;
3 Hong Kong Polytechnic University, Hong Kong, China
1590
Objectives: Functional imaging is valuable for the early diagnosis and classification of dementia. However, accurate image interpretation and classifying the particular dementia syndrome remains an important challenge. Our aim was to develop a multiclass probabilistic algorithm using data mining to separate Alzheimer’s disease (AD), fronto-temporal dementia (FTD) and normals using FDG-PET parametric images.
Methods: Parametric images of rCMRGlc were spatially normalized using SPM2. Voxels were down-sampled for each dimension by a factor of 1.5, followed by further dimension reduction using principal component analysis (PCA). The support vector machine (SVM) method was used in data mining with a radial basis function kernel. PCA with 5, 12 and 50 components were investigated. Ten-fold cross validation and leave-one-out cross validation were applied to obtain optimum parameters for SVM. Fuzzy dementia classifications were then achieved using the obtained SVM model.
Results: 235 neurological studies were analysed. Patients were referred from dementia clinics and images were interpreted by an experienced PET physician/neurologist. The image interpretation was AD in 90, FTD in 77 and normal in 68. The highest accuracy was achieved by SVM with 12 PCA components, whose results are compared with the original classifications in the table.
Conclusions: Our results show that data mining with SVM and PCA for dimension reduction shows potential in providing fuzzy classification of multiclass dementia and hence may aid the diagnosis and classification of the dementing syndromes.
