Recognition of Alzheimer's disease and Mild Cognitive Impairment with multimodal image-derived biomarkers and Multiple Kernel Learning

详细信息    查看全文

• 作者：Olfa Ben Ahmed^a ; ^{olfa.ben-ahmed@unilim.fr}Author Vitae ; Jenny Benois-Pineau^b ; ^{jenny.benois@labri.fr}Author Vitae ; Michelle Allard^c ; ^d ; ^e ; ^{michelle.allard@chu-bordeaux.fr}Author Vitae ; Gwé ; naë ; lle Catheline^c ; ^d ; ^e ; ^{gwenaelle.catheline@chu-bordeaux.fr}Author Vitae ; Chokri Ben Amar^f ; ^{chokri.benamar@ieee.org}Author Vitae ; For the Alzheimer's Disease Neuroimaging Initiative¹
• 关键词：Alzheimer's disease ; Multiple Kernel Learning ; Mutlimodal fusion ; Diagnosis ; Local features ; DTI ; MD maps ; CHFs ; Imaging biomarkers
• 刊名：Neurocomputing
• 出版年：2017
• 出版时间：12 January 2017
• 年：2017
• 卷：220
• 期：Complete
• 页码：98-110
• 全文大小：2176 K
• 卷排序：220

文摘

Computer-Aided Diagnosis (CAD) of Alzheimer's disease (AD) has drawn the attention of computer vision research community over the last few years. Several attempts have been made to adapt pattern recognition approaches to specific neuroimaging data such as Structural MRI (sMRI) for early AD diagnosis. One strategy is to boost the discrimination power of such approaches by integrating complementary imaging modalities in a single learning framework. Diffusion Tensor Imaging (DTI) is a new and promising modality giving complementary information to the anatomical MRI. However, including relevant DTI information from such modality is a challenging problem. In this paper, we propose to extract local image-derived biomarkers from DTI and sMRI to construct multimodal AD signatures. To assess the relevance of such modalities as well as to optimize the classifier, we integrate complementary information using a Multiple Kernel Learning (MKL) framework for AD subjects recognition. To evaluate our method, we perform experiments on a subset from the Alzheimer's Disease Neuroimaging Initiative (ADNI) dataset. Both T1-weighted MRI and Mean Diffusivity (MD) maps from the DTI modality of 45 AD patients, 52 Normal Control (NC) and 58 Mild Cognitive Impairment (MCI) subjects have been used. The obtained results indicate that our multimodal approach yields significant improvement in accuracy over using each single modality independently. The classification accuracies obtained by the proposed method are 90.2%, 79.42% and 76.63% for respectively AD vs. NC, MCI vs. NC and AD vs. MCI binary classification problems. For the MCI classification problem, the proposed fusion framework leads to an average increase about at least 9% for the accuracy, 5% for the specificity and 15% for the sensitivity.