A fuzzy-based system reveals Alzheimer's Disease onset in subjects with Mild Cognitive Impairment. (June 2017)

• Record Type: Journal Article
• Authors: Tangaro, Sabina; Fanizzi, Annarita; Amoroso, Nicola; Bellotti, Roberto
• Is Part Of: Physica medica. Volume 38(2017)
• Page Start: 36
• Page End: 44
• Publication Date: 2017-06
• Subjects: Fuzzy logic -- Alzheimer's Disease -- MCI -- Early diagnosis -- MRI -- Cognitive measurements -- Support Vector Machine; Medical physics -- Periodicals; Biophysics -- Periodicals; Biophysics -- Periodicals; Imagerie médicale -- Périodiques; Radiothérapie -- Périodiques; Rayons X -- Sécurité -- Mesures -- Périodiques; Physique -- Périodiques; Médecine -- Périodiques; 610.153
• DOI: 10.1016/j.ejmp.2017.04.027
• Languages: English
• ISSNs: 1120-1797

Abstract

Highlights: A prediction model of conversion to Alzheimer's Disease is proposed. The model is trained on the structural features of MRI and cognitive measurements. The fuzzy classes of the hippocampal volume have driven classification. The model accurately predicted the disease onset also after one year. The fuzzy approach is robust and can find a wide range of applications. Abstract: Alzheimer's Disease (AD) is the most frequent neurodegenerative form of dementia. Although dementia cannot be cured, it is very important to detect preclinical AD as early as possible. Several studies demonstrated the effectiveness of the joint use of structural Magnetic Resonance Imaging (MRI) and cognitive measures to detect and track the progression of the disease. Since hippocampal atrophy is a well known biomarker for AD progression state, we propose here a novel methodology, exploiting it as a searchlight to detect the best discriminating features for the classification of subjects with Mild Cognitive Impairment (MCI) converting (MCI-c) or not converting (MCI-nc) to AD. In particular, we define a significant subdivision of the hippocampal volume in fuzzy classes, and we train for each class Support Vector Machine SVM classifiers on cognitive and morphometric measurements of normal controls (NC) and AD patients. From the ADNI database, we used MRI scans and cognitive measurements at baseline of 372 subjects, including 98 subjects with AD, and 117 NC as a training set, 86 with MCI-c and 71 with MCI-nc as an independent test set. The accuracy of early diagnosis was evaluated by means of a longitudinal analysis. The proposed methodology was able to accurately predict the disease onset also after one year (median AUC = 88.2%, interquartile range 87.2%–89.0%). Besides its robustness, the proposed fuzzy methodology naturally incorporates the uncertainty degree intrinsically affecting neuroimaging features. Thus, it might be applicable in several other pathological conditions affecting morphometric changes of the brain.