Characterization of brain anatomical patterns by comparing region intensity distributions: Applications to the description of Alzheimer's disease

Abstract

PurposeThis work presents an automatic characterization of the Alzheimer's disease describing the illness as a multidirectional departure from a baseline defining the control state, being these directions determined by a distance between functional‐equivalent anatomical regions.MethodsAfter a brain parcellation, a region is described by its histogram of gray levels, and the Earth mover's distance establishes how close or far these regions are. The medoid of the control group is set as the reference and any brain is characterized by its set of distances to this medoid.EvaluationThis hypothesis was assessed by separating groups of patients with mild Alzheimer's disease and mild cognitive impairment from control subjects, using a subset of the Open Access Series of Imaging Studies (OASIS) database. An additional experiment evaluated the method generalization and consisted in training with the OASIS data and testing with the Minimal Interval Resonance Imaging in Alzheimer's disease (MIRIAD) database.ResultsClassification between controls and patients with AD resulted in an equal error rate of 0.1 (90% of sensitivity and specificity at the same time). The automatic ranking of regions resulting is in strong agreement with those regions described as important in clinical practice. Classification with different databases results in a sensitivity of 85% and a specificity of 91%.ConclusionsThis method automatically finds out a multidimensional expression of the AD, which is directly related to the anatomical changes in specific areas such as the hippocampus, the amygdala, the planum temporale, and thalamus.

Applying deep learning models to structural MRI for stage prediction of Alzheimer’s disease

“…The proposed CNN model has the best performance in terms of accuracy values for diagnosis of MCI. The other study [11] reported a 0.74 AUC value, which is higher than the 0.6 AUC value of the proposed CNN model. However, the feature engineering applied in the other study [11] should be evaluated carefully while comparing the performance of the two studies.…”

“…They did not provide an accuracy value for model performance. Although the proposed CNN model has a lower AUC value (0.85) compared to the study by Giraldo et al [11] (0.92), the proposed model does not explicitly extract MRI features (i.e. brain parcellation process on MRI); instead it uses automatically computed features provided by convolutional and pooling layers of the CNN model.…”

“…In the current study, the proposed model demonstrated a higher accuracy with 83% on the whole brain images. Alzheimer's (MRI) MIRIAD -0.84 0.85 0.83 Li et al [10] Alzheimer's (MRI) OASIS -0.75 0.74 0.78 Giraldo et al [11] Alzheimer's (MRI) OASIS+MIRIAD 0.92 -0.90 0.90 Zhang et al [26] Alzheimer's (MRI) ADNI -0.85 --Ahmed et al [27] Alzheimer's (MRI) ADNI -0.84 0.79 0.88 Ahmed et al [27] Alzheimer's (MRI) BORDEAUX-3CITY -0.78 0.80 0.75 Demirhan [28] Alzheimer A limited number of studies in the literature use separate data sets for training and evaluation purposes. In the current study, the OASIS data set was used only for the training stage and the MIRIAD data set used only in the evaluation stage.…”

“…In the current study, the OASIS data set was used only for the training stage and the MIRIAD data set used only in the evaluation stage. This setup was not applied in recent studies except for in the study by Giraldo et al [11], who used a special method to extract brain regions, which were represented by histograms of gray levels, and the differences between subjects were determined by using Earth mover's distance. They employed a Random Unit Sampling Boost method to perform diagnosis of patients.…”

“…Histogram-based features obtained from MRI data were applied to classify different brain levels by using an SVM. A recent study provided a method for automatically detecting anatomical changes in the hippocampus, amygdala, planum temporale, and thalamus regions [11]. Differences and similarities between two individuals were identified according to gray level histogram values on MRI data.…”