Multimodal deep learning for Alzheimer's disease dementia assessment.

Worldwide, there are nearly 10 million new cases of dementia annually, of which Alzheimer's disease (AD) is the most common. New measures are needed to improve the diagnosis of individuals with cognitive impairment due to various etiologies. Here, we report a deep learning framework that accomplishes multiple diagnostic steps in successive fashion to identify persons with normal cognition (NC), mild cognitive impairment (MCI), AD, and non-AD dementias (nADD). We demonstrate a range of models capable of accepting flexible combinations of routinely collected clinical information, including demographics, medical history, neuropsychological testing, neuroimaging, and functional assessments. We then show that these frameworks compare favorably with the diagnostic accuracy of practicing neurologists and neuroradiologists. Lastly, we apply interpretability methods in computer vision to show that disease-specific patterns detected by our models track distinct patterns of degenerative changes throughout the brain and correspond closely with the presence of neuropathological lesions on autopsy. Our work demonstrates methodologies for validating computational predictions with established standards of medical diagnosis.

Fully automated intracranial ventricle segmentation on CT with 2D regional convolutional neural network to estimate ventricular volume.

PURPOSE: Hydrocephalus is a clinically significant condition which can have devastating consequences if left untreated. Currently available methods for quantifying this condition using CT imaging are unreliable and prone to error. The purpose of this study is to investigate the clinical utility of using convolutional neural networks to calculate ventricular volume and explore limitations. METHODS: A two-dimensional convolutional neural network was designed to perform fully automated ventricular segmentation on CT images. A total of 300 head CTs were collected and used in this exploration. Two hundred were used to train the network, 50 were used for validation, and 50 were used for testing. RESULTS: Dice scores for the left lateral, right lateral, and third ventricle segmentations were 0.92, 0.92, and 0.79, respectively; the coefficients of determination were r2 = 0.991, r2 = 0.994, and r2 = 0.976; the average volume differences between manual and automated segmentation were 0.821 ml, 0.587 ml, and 0.099 ml. CONCLUSION: Two-dimensional convolutional neural network architectures can be used to accurately segment and quantify intracranial ventricle volume. While further refinements are necessary, it is likely these networks could be used as a clinical tool to quantify hydrocephalus accurately and efficiently.

Reliability and reproducibility of individual differences in functional connectivity acquired during task and resting state.

OBJECTIVES: Application of fMRI connectivity metrics as diagnostic biomarkers at the individual level will require reliability, sensitivity and specificity to longitudinal changes in development, aging, neurocognitive, and behavioral performance and pathologies. Such metrics have not been well characterized for recent advances in BOLD acquisition. EXPERIMENTAL DESIGN: Analysis of multiband BOLD data from the HCP 500 Subjects Release was performed with FIX ICA and with WM, CSF and motion parameter regression. Analysis with ROIs covering the gray matter at 5 mm resolution was performed to assess functional connectivity. ROIs in key areas were used to demonstrate statistical differences between specific connections. Reproducibility of group-mean functional connectivity and for single connections for individuals was evaluated for both resting state and task acquisitions. PRINCIPAL OBSERVATIONS: Systematic differences in group-mean connectivity were demonstrated during task and rest and during different tasks, although individual differences in connectivity were maintained. Reproducibility of a single connection for a subject and across subjects for resting and task acquisition was demonstrated to be a linear function of the square root of imaging time. Randomly removing up to 50% of time points had little effect on reliability, while truncating an acquisition was associated with decreased reliability. Reliability was highest within the cortex, and lowest for deep gray nuclei, gray-white junction, and near large sulci. CONCLUSIONS: This study found systematic differences in group-mean connectivity acquired during task and rest acquitisions and preserved individual differences in connectivity due to intrinsic differences in an individual's brain activity and structural brain architecture. We also show that longer scan times are needed to acquire data on single subjects for information on connections between specific ROIs. Longer scans may be facilitated by acquisition during task paradigms, which will systematically affect functional connectivity but may preserve individual differences in connectivity on top of task modulations.