Maciej Plocharski1, Lasse Riis Østergaard2. 1. Department of Health Science and Technology, Aalborg University, Aalborg, Denmark. Electronic address: mpl@hst.aau.dk. 2. Department of Health Science and Technology, Aalborg University, Aalborg, Denmark.
Abstract
BACKGROUND: Recent advancements in medical imaging have resulted in a significant growth in diagnostic possibilities of neurodegenerative disorders. Neuroanatomical abnormalities of the cerebral cortex in Alzheimer's disease (AD), the most frequent type of dementia in the elderly, can be observed in morphology analysis of cortical sulci, and used to distinguish between cognitively normal (CN) subjects and subjects with AD. OBJECTIVE: The purpose of this paper was to extract sulcal features by means of computing a sulcal medial surface for AD/CN classification. METHODS: 24 distinct sulci per subject were extracted from 210 subjects from the ADNI database by the BrainVISA sulcal identification pipeline. Sulcal medial surface features (depth, length, mean and Gaussian curvature, surface area) were computed for AD/CN classification with a support vector machine (SVM). RESULTS: The obtained 10-fold cross-validated classification accuracy was 87.9%, sensitivity 90.0%, and specificity 86.7%, based on ten features. The area under the receiver operating characteristic curve (AUC) was 0.89. CONCLUSIONS: The sulcal medial surface features can be used as biomarkers for cortical neuroanatomical abnormalities in AD. All the features were located in the left hemisphere, which had previously been reported to be more severely affected in AD and to lose grey matter faster than the right hemisphere.
BACKGROUND: Recent advancements in medical imaging have resulted in a significant growth in diagnostic possibilities of neurodegenerative disorders. Neuroanatomical abnormalities of the cerebral cortex in Alzheimer's disease (AD), the most frequent type of dementia in the elderly, can be observed in morphology analysis of cortical sulci, and used to distinguish between cognitively normal (CN) subjects and subjects with AD. OBJECTIVE: The purpose of this paper was to extract sulcal features by means of computing a sulcal medial surface for AD/CN classification. METHODS: 24 distinct sulci per subject were extracted from 210 subjects from the ADNI database by the BrainVISA sulcal identification pipeline. Sulcal medial surface features (depth, length, mean and Gaussian curvature, surface area) were computed for AD/CN classification with a support vector machine (SVM). RESULTS: The obtained 10-fold cross-validated classification accuracy was 87.9%, sensitivity 90.0%, and specificity 86.7%, based on ten features. The area under the receiver operating characteristic curve (AUC) was 0.89. CONCLUSIONS: The sulcal medial surface features can be used as biomarkers for cortical neuroanatomical abnormalities in AD. All the features were located in the left hemisphere, which had previously been reported to be more severely affected in AD and to lose grey matter faster than the right hemisphere.