Quantifying brain metabolism from FDG-PET images into a probability of Alzheimer's dementia score.

18 F-fluorodeoxyglucose positron emission tomography (FDG-PET) enables in-vivo capture of the topographic metabolism patterns in the brain. These images have shown great promise in revealing the altered metabolism patterns in Alzheimer's disease (AD). The AD pathology is progressive, and leads to structural and functional alterations that lie on a continuum. There is a need to quantify the altered metabolism patterns that exist on a continuum into a simple measure. This work proposes a 3D convolutional neural network with residual connections that generates a probability score useful for interpreting the FDG-PET images along the continuum of AD. This network is trained and tested on images of stable normal control and stable Dementia of the Alzheimer's type (sDAT) subjects, achieving an AUC of 0.976 via repeated fivefold cross-validation. An independent test set consisting of images in between the two extreme ends of the DAT spectrum is used to further test the generalization performance of the network. Classification performance of 0.811 AUC is achieved in the task of predicting conversion of mild cognitive impairment to DAT for conversion time of 0-3 years. The saliency and class activation maps, which highlight the regions of the brain that are most important to the classification task, implicate many known regions affected by DAT including the posterior cingulate cortex, precuneus, and hippocampus.

3D hemisphere-based convolutional neural network for whole-brain MRI segmentation.

Whole-brain segmentation is a crucial pre-processing step for many neuroimaging analyses pipelines. Accurate and efficient whole-brain segmentations are important for many neuroimage analysis tasks to provide clinically relevant information. Several recently proposed convolutional neural networks (CNN) perform whole brain segmentation using individual 2D slices or 3D patches as inputs due to graphical processing unit (GPU) memory limitations, and use sliding windows to perform whole brain segmentation during inference. However, these approaches lack global and spatial information about the entire brain and lead to compromised efficiency during both training and testing. We introduce a 3D hemisphere-based CNN for automatic whole-brain segmentation of T1-weighted magnetic resonance images of adult brains. First, we trained a localization network to predict bounding boxes for both hemispheres. Then, we trained a segmentation network to segment one hemisphere, and segment the opposing hemisphere by reflecting it across the mid-sagittal plane. Our network shows high performance both in terms of segmentation efficiency and accuracy (0.84 overall Dice similarity and 6.1 mm overall Hausdorff distance) in segmenting 102 brain structures. On multiple independent test datasets, our method demonstrated a competitive performance in the subcortical segmentation task and a high consistency in volumetric measurements of intra-session scans.

Construction of MRI-Based Alzheimer's Disease Score Based on Efficient 3D Convolutional Neural Network: Comprehensive Validation on 7,902 Images from a Multi-Center Dataset.

BACKGROUND: In recent years, many convolutional neural networks (CNN) have been proposed for the classification of Alzheimer's disease. Due to memory constraints, many of the proposed CNNs work at a 2D slice-level or 3D patch-level. OBJECTIVE: Here, we propose a subject-level 3D CNN that can extract the neurodegenerative patterns of the whole brain MRI and converted into a probabilistic Dementia score. METHODS: We propose an efficient and lightweight subject-level 3D CNN featuring dilated convolutions. We trained our network on the ADNI data on stable Dementia of the Alzheimer's type (sDAT) from stable normal controls (sNC). To comprehensively evaluate the generalizability of our proposed network, we performed four independent tests which includes testing on images from other ADNI individuals at various stages of the dementia, images acquired from other sites (AIBL), images acquired using different protocols (OASIS), and longitudinal images acquired over a short period of time (MIRIAD). RESULTS: We achieved a 5-fold cross-validated balanced accuracy of 88%in differentiating sDAT from sNC, and an overall specificity of 79.5%and sensitivity 79.7%on the entire set of 7,902 independent test images. CONCLUSION: Independent testing is essential for estimating the generalization ability of the network to unseen data, but is often lacking in studies using CNN for DAT classification. This makes it difficult to compare the performances achieved using different architectures. Our comprehensive evaluation highlighting the competitive performance of our network and potential promise for generalization.

Blinded Clinical Evaluation for Dementia of Alzheimer's Type Classification Using FDG-PET: A Comparison Between Feature-Engineered and Non-Feature-Engineered Machine Learning Methods.

BACKGROUND: Advanced machine learning methods can aid in the identification of dementia risk using neuroimaging-derived features including FDG-PET. However, to enable the translation of these methods and test their usefulness in clinical practice, it is crucial to conduct independent validation on real clinical samples, which has yet to be properly delineated in the current literature. OBJECTIVE: In this paper, we present our efforts to enable such clinical translational through the evaluation and comparison of two machine-learning methods for discrimination between dementia of Alzheimer's type (DAT) and Non-DAT controls. METHODS: FDG-PET-based dementia scores were generated on an independent clinical sample whose clinical diagnosis was blinded to the algorithm designers. A feature-engineered approach (multi-kernel probability classifier) and a non-feature-engineered approach (3D convolutional neural network) were analyzed. Both classifiers were pre-trained on cognitively normal subjects as well as subjects with DAT. These two methods provided a probabilistic dementia score for this previously unseen clinical data. Performance of the algorithms were compared against ground-truth dementia rating assessed by experienced nuclear physicians. RESULTS: Blinded clinical evaluation on both classifiers showed good separation between the cognitively normal subjects and the patients diagnosed with DAT. The non-feature-engineered dementia score showed higher sensitivity among subjects whose diagnosis was in agreement between the machine-learning models, while the feature-engineered approach showed higher specificity in non-consensus cases. CONCLUSION: In this study, we demonstrated blinded evaluation using data from an independent clinical sample for assessing the performance in DAT classification models in a clinical setting. Our results showed good generalizability for two machine-learning approaches, marking an important step for the translation of pre-trained machine-learning models into clinical practice.