Analysis of Alzheimer’s Disease Based on the Random Neural Network Cluster in fMRI

Using functional magnetic resonance imaging (fMRI) we have evaluated the anatomical location of the motor hand area. The segment of the precentral gyrus that most often contained motor hand function was a knob-like structure, that is shaped like an omega or epsilon in the axial plane and like a hook in the sagittal plane. On the cortical surface of cadaver specimens this precentral knob corresponded precisely to the characteristic 'middle knee' of the central sulcus that has been described by various anatomists in the last century. We were then able to show that this knob is a reliable landmark for identifying the precentral gyrus directly. We therefore conclude that neural elements involved in motor hand function are located in a characteristic 'precentral knob' which is a reliable landmark for identifying the precentral gyrus under normal and pathological conditions. It faces and forms the 'middle knee' of the central sulcus, is located just at the cross point between the precentral sulcus and the central sulcus, and is therefore also visible on the cortical surface.

The superior frontal gyrus (SFG) is thought to contribute to higher cognitive functions and particularly to working memory (WM), although the nature of its involvement remains a matter of debate. To resolve this issue, methodological tools such as lesion studies are needed to complement the functional imaging approach. We have conducted the first lesion study to investigate the role of the SFG in WM and address the following questions: do lesions of the SFG impair WM and, if so, what is the nature of the WM impairment? To answer these questions, we compared the performance of eight patients with a left prefrontal lesion restricted to the SFG with that of a group of 11 healthy control subjects and two groups of patients with focal brain lesions [prefrontal lesions sparing the SFG (n = 5) and right parietal lesions (n = 4)] in a series of WM tasks. The WM tasks (derived from the classical n-back paradigm) allowed us to study the impact of the SFG lesions on domain (verbal, spatial, face) and complexity (1-, 2- and 3-back) processing within WM. As expected, patients with a left SFG lesion exhibited a WM deficit when compared with all control groups, and the impairment increased with the complexity of the tasks. This complexity effect was significantly more marked for the spatial domain. Voxel-to-voxel mapping of each subject's performance showed that the lateral and posterior portion of the SFG (mostly Brodmann area 8, rostral to the frontal eye field) was the subregion that contributed the most to the WM impairment. These data led us to conclude that (i) the lateral and posterior portion of the left SFG is a key component of the neural network of WM; (ii) the participation of this region in WM is triggered by the highest level of executive processing; (iii) the left SFG is also involved in spatially oriented processing. Our findings support a hybrid model of the anatomical and functional organization of the lateral SFG for WM, according to which this region is involved in higher levels of WM processing (monitoring and manipulation) but remains oriented towards spatial cognition, although the domain specificity is not exclusive and is overridden by an increase in executive demand, regardless of the domain being processed. From a clinical perspective, this study provides new information on the impact of left SFG lesions on cognition that will be of use to neurologists and neurosurgeons.

For the last decade, it has been shown that neuroimaging can be a potential tool for the diagnosis of Alzheimer's Disease (AD) and its prodromal stage, Mild Cognitive Impairment (MCI), and also fusion of different modalities can further provide the complementary information to enhance diagnostic accuracy. Here, we focus on the problems of both feature representation and fusion of multimodal information from Magnetic Resonance Imaging (MRI) and Positron Emission Tomography (PET). To our best knowledge, the previous methods in the literature mostly used hand-crafted features such as cortical thickness, gray matter densities from MRI, or voxel intensities from PET, and then combined these multimodal features by simply concatenating into a long vector or transforming into a higher-dimensional kernel space. In this paper, we propose a novel method for a high-level latent and shared feature representation from neuroimaging modalities via deep learning. Specifically, we use Deep Boltzmann Machine (DBM)(2), a deep network with a restricted Boltzmann machine as a building block, to find a latent hierarchical feature representation from a 3D patch, and then devise a systematic method for a joint feature representation from the paired patches of MRI and PET with a multimodal DBM. To validate the effectiveness of the proposed method, we performed experiments on ADNI dataset and compared with the state-of-the-art methods. In three binary classification problems of AD vs. healthy Normal Control (NC), MCI vs. NC, and MCI converter vs. MCI non-converter, we obtained the maximal accuracies of 95.35%, 85.67%, and 74.58%, respectively, outperforming the competing methods. By visual inspection of the trained model, we observed that the proposed method could hierarchically discover the complex latent patterns inherent in both MRI and PET.