{"title":"3D Multi-feature fusion convolutional network for Alzheimer's disease diagnosis.","authors":"Jiao Jiao Feng, Mao Wen Ba, Nan Li, Gang Wang","doi":"10.1109/EMBC53108.2024.10782006","DOIUrl":null,"url":null,"abstract":"<p><p>The cognitive decline caused by Alzheimer's disease (AD) is closely related to the structural changes in the hippocampus captured by structural magnetic resonance imaging (sMRI). However, current deep model research on the morphological analysis of hippocampus is mainly based on 2D MRI slices, lacking a comprehensive description of the 3D surface morphology and complex textures of the hippocampus. For this reason, we propose a two-stream multi features deep learning model that establishes a descriptive system for 3D spatial structure and morphological atrophy features on the triangular mesh of left and right hippocampus. First, we encode the triangular mesh data into the spatial structural features of the hippocampal surface. Second, considering the tubular structure of the hippocampus and the inhomogeneous morphological changes caused by AD, we introduce the thickness features and Heat Kernel Signature (HKS) features for the morphological atrophy features encoding. Third, we integrate the encoded features of adjacent faces from a macroscopic perspective into the discriminative morphological features induced by AD. Finally, driven by classification tasks, the deep learning model parameters and the discriminative features are continuously optimized, thereby improving the accuracy of AD diagnosis. Our method is evaluated based on the T 1 weighted sMRI baseline data of 269 Aβ+ AD and 437 Aβ-normal cognitively(NC) subjects collected from the Alzheimer's Disease Neuroimaging Initiative (ADNI) database. The classification accuracy of this method for AD and NC subjects is 93.4%, the sensitivity and specificity are 92.4% and 93.8%, respectively, and the area under the ROC curve (AUC) is 98.3%.</p>","PeriodicalId":72237,"journal":{"name":"Annual International Conference of the IEEE Engineering in Medicine and Biology Society. IEEE Engineering in Medicine and Biology Society. Annual International Conference","volume":"2024 ","pages":"1-6"},"PeriodicalIF":0.0000,"publicationDate":"2024-07-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":"0","resultStr":null,"platform":"Semanticscholar","paperid":null,"PeriodicalName":"Annual International Conference of the IEEE Engineering in Medicine and Biology Society. IEEE Engineering in Medicine and Biology Society. Annual International Conference","FirstCategoryId":"1085","ListUrlMain":"https://doi.org/10.1109/EMBC53108.2024.10782006","RegionNum":0,"RegionCategory":null,"ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":null,"EPubDate":"","PubModel":"","JCR":"","JCRName":"","Score":null,"Total":0}
引用次数: 0
Abstract
The cognitive decline caused by Alzheimer's disease (AD) is closely related to the structural changes in the hippocampus captured by structural magnetic resonance imaging (sMRI). However, current deep model research on the morphological analysis of hippocampus is mainly based on 2D MRI slices, lacking a comprehensive description of the 3D surface morphology and complex textures of the hippocampus. For this reason, we propose a two-stream multi features deep learning model that establishes a descriptive system for 3D spatial structure and morphological atrophy features on the triangular mesh of left and right hippocampus. First, we encode the triangular mesh data into the spatial structural features of the hippocampal surface. Second, considering the tubular structure of the hippocampus and the inhomogeneous morphological changes caused by AD, we introduce the thickness features and Heat Kernel Signature (HKS) features for the morphological atrophy features encoding. Third, we integrate the encoded features of adjacent faces from a macroscopic perspective into the discriminative morphological features induced by AD. Finally, driven by classification tasks, the deep learning model parameters and the discriminative features are continuously optimized, thereby improving the accuracy of AD diagnosis. Our method is evaluated based on the T 1 weighted sMRI baseline data of 269 Aβ+ AD and 437 Aβ-normal cognitively(NC) subjects collected from the Alzheimer's Disease Neuroimaging Initiative (ADNI) database. The classification accuracy of this method for AD and NC subjects is 93.4%, the sensitivity and specificity are 92.4% and 93.8%, respectively, and the area under the ROC curve (AUC) is 98.3%.