Identifying the regional substrates predictive of Alzheimer's disease progression through a convolutional neural network model and occlusion

Abstract

Progressive brain atrophy is a key neuropathological hallmark of Alzheimer's disease (AD) dementia. However, atrophy patterns along the progression of AD dementia are diffuse and variable and are often missed by univariate methods. Consequently, identifying the major regional atrophy patterns underlying AD dementia progression is challenging. In the current study, we propose a method that evaluates the degree to which specific regional atrophy patterns are predictive of AD dementia progression, while holding all other atrophy changes constant using a total sample of 334 subjects.We first trained a dense convolutional neural network model to differentiate individuals with mild cognitive impairment (MCI) who progress to AD dementia versus those with a stable MCI diagnosis. Then, we retested the model multiple times, each time occluding different regions of interest (ROIs) from the model's testing set's input. We also validated this approach by occluding ROIs based on Braak's staging scheme. We found that the hippocampus, fusiform, and inferior temporal gyri were the strongest predictors of AD dementia progression, in agreement with established staging models. We also found that occlusion of limbic ROIs defined according to Braak stage III had the largest impact on the performance of the model. Our predictive model reveals the major regional patterns of atrophy predictive of AD dementia progression.These results highlight the potential for early diagnosis and stratification of individuals with prodromal AD dementia based on patterns of cortical atrophy, prior to interventional clinical trials.

Fused Sparse Network Learning for Longitudinal Analysis of Mild Cognitive Impairment

“…Recently, some studies have shown that models trained by NC and AD outperform those trained by sMCI and pMCI for predicting the progression of MCI [4], [8], [7], [9], [10], [23]. Huang et al proposed a novel CNN to fuse the multi-modality information from T1-MRI and FDG-PET images around the hippocampus area for predicting MCI progression [4].…”

“…However, prospectively acquiring MCI subtype data is time-consuming and resource-intensive, which leads to relatively small labeled datasets, resulting in amplified overfitting and challenges in extracting discriminative information for real-world scenarios. Recently, some previous studies revealed that the models trained by NC and AD data perform better than those trained by sMCI and pMCI data for predicting the progression of MCI [4], [8], [9], [10]. However, these studies only considered that differences between sMCI and pMCI are analogical to those between NC and AD, ignoring the global ordinal nature of AD progression.…”

The spatial distribution of coupling between tau and neurodegeneration in amyloid-β positive mild cognitive impairment

“…Early involvement of these regions in AD pathogenesis is well documented (35)(36)(37). Moreover, atrophy in the hippocampus and adjacent early Braak regions is predictive of conversion from MCI to AD (38,39). Association between tau and atrophy in regions corresponding to early Braak stages are also evident in Aβ-CN individuals (40), however the current results demonstrate that coupling was stronger when Aβ load was greater.…”

Deep Learning for Brain MRI Confirms Patterned Pathological Progression in Alzheimer's Disease

“…Here, the 531 of the 638 sMRI images of AD subjects are selected for the experiment, because the remaining 107 images have no corresponding MMSE values. We explore the frequent item-sets consisting of the neurodegenerative regions in each level (normal: MMSE ∈ [27,30]; mild: MMSE ∈ [21,26]; moderate: MMSE ∈ [10,20]; severe: MMSE ∈ [0, 9]) to analyze the process of neurodegeneration at different stages of AD. As summarized in Table 7, in the early stage of cognitive impairment (MMSE ∈ [27,30]), L.mAmyg and L.NAC have been neurodegenerative in most (over 83%) AD sMRI images.…”

“…[19] The situation might be a reason why little attention has been paid to analyzing patterned pathological progression in AD based on DL models rather than targeting diagnostic classification tasks. [20][21][22][23] Some studies have started to extract atrophy features [24] or patterns [25] from sMRI to derive the brain age. In line with this effort, our recent study [26] has developed a DL model (2-dimensional convolutional neural network) to identify the critical discriminative brain regions for AD recognition.…”