Search

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.

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.

Optical coherence tomography angiography (OCT-A) may help in the diagnosis of diabetic retinopathy (DR). The OCT-A field-of-view and the volume acquisition rate are limited by the number of samples. We explore the effect of reduced lateral sampling on DR classification with Deep Learning. © 2021 The Authors.

PurposeInvestigate the effects of myopia and glaucoma in the prelaminar neural canal and anterior lamina cribrosa using 1060-nm swept-source optical coherence tomographyDesignRetrospective, cross-sectional studyMethodsSettingClinical practicePatient or study population19 controls (38 eyes); 38 glaucomatous subjects (63 eyes). Inclusion criteria for glaucomatous subjects: i) optic disc neural rim loss; ii) peripapillary nerve fibre layer (NFL) loss on spectral domain optical coherence tomography (SD-OCT); iii) glaucomatous visual field defect with an abnormal pattern standard deviation (PIntervention or observation proceduresSwept-source optical coherence tomographyMain Outcome Measure(s)Bruchs membrane opening (BMO) and anterior laminar insertion (ALI) dimension, prelaminar neural canal dimension, anterior lamina cribrosa surface (ALCS) depthResultsGlaucomatous eyes had more bowed and nasally rotated BMO and ALI, more horizontally skewed prelaminar neural canal, and deeper ALCS than the control eyes. Increased axial length was associated with a wider, longer, and more horizontally skewed neural canal, and decrease in the ALCS depth and curvature.ConclusionOur findings suggest that glaucomatous posterior bowing or cupping of lamina cribrosa can be significantly confounded by the myopic expansion of the neural canal. This may be related to higher glaucoma risk associated with myopia from decreased compliance and increased susceptibility to IOP-related damage of LC being pulled taut.

Megahertz-rate optical coherence tomography angiography (OCTA) is highly anticipated as an ultrafast imaging tool in clinical settings. However, shot-noise-limited sensitivity is inevitably reduced in high-speed imaging systems. In this Letter, we present a coherent buffer averaging technique for use with a Fourier-domain mode-locked (FDML) laser to improve OCTA contrast at 1060 nm MHz-rate retinal imaging. Full characterization of spectral variations among the FDML buffers and a numerical correction method are also presented, with the results demonstrating a 10-fold increase in the phase alignment among buffers. Coherent buffer averaging provided better OCTA contrast than the conventional multi-frame averaging approach with a faster acquisition time.

Optical coherence tomography (OCT) and OCT angiography (OCT-A) may benefit the screening of diabetic retinopathy (DR). This study investigated the effect of laterally subsampling OCT/OCT-A en face scans by up to a factor of 8 when using deep neural networks for automated referable DR classification. There was no significant difference in the classification performance across all evaluation metrics when subsampling up to a factor of 3, and only minimal differences up to a factor of 8. Our findings suggest that OCT/OCT-A can reduce the number of samples (and hence the acquisition time) for a volume for a given field of view on the retina that is acquired for rDR classification.

PReCIS:: The Bruch membrane opening (BMO) was posteriorly bowed and the degree of nonplanarity increased in stable and progressive glaucoma subjects. BMO became more posterior relative to the Bruch membrane (BM) in control and both stable and progressive glaucoma subjects. Purpose To investigate longitudinal changes in morphologic characteristics of the BMO in control and glaucomatous subjects. Materials and methods A total of 53 myopic eyes (17 control, 6 suspect, 20 stable glaucoma, and 10 progressing glaucoma) were followed for an average of 4.2±1.4 years and imaged at the baseline and 2 follow-up appointments using a 1060 nm swept-source optical coherence tomography system. BM and BMO were segmented, and 4 morphometric BMO parameters (area, ellipse ratio, nonplanarity, and depth) were measured. Results There were no significant changes in BMO area or ellipse ratio for all groups. BMO nonplanarity was shown to increase in the glaucoma groups. BMO depth relative to BM increased in all groups except the suspects (control: 8.1 µm/y, P=0.0001; stable glaucoma: 3.5 µm/y, P=0.0001; progressing glaucoma: 14.0 µm/y, P=0.0026). In linear mixed-model analysis, axial length was positively associated with BMO area in all groups except for progressing glaucoma, and with BMO nonplanarity in stable glaucoma. It was not a significant factor to the slopes of the BMO parameters in the ANCOVA analysis of slopes. Conclusions Longitudinally, BMO increased in nonplanarity in the glaucoma eyes, and its axial position relative to BM became more posterior in both control and glaucoma eyes.

18F‐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.

Retinal swelling due to the accumulation of fluid is associated with the most vision-threatening retinal diseases. Optical coherence tomography (OCT) is the current standard of care in assessing the presence and quantity of retinal fluid and image-guided treatment management. Deep learning methods have made their impact across medical imaging, and many retinal OCT analysis methods have been proposed. However, it is currently not clear how successful they are in interpreting the retinal fluid on OCT, which is due to the lack of standardized benchmarks. To address this, we organized a challenge RETOUCH in conjunction with MICCAI 2017, with eight teams participating. The challenge consisted of two tasks: fluid detection and fluid segmentation. It featured for the first time: all three retinal fluid types, with annotated images provided by two clinical centers, which were acquired with the three most common OCT device vendors from patients with two different retinal diseases. The analysis revealed that in the detection task, the performance on the automated fluid detection was within the inter-grader variability. However, in the segmentation task, fusing the automated methods produced segmentations that were superior to all individual methods, indicating the need for further improvements in the segmentation performance.

In diseases such as cancer, patients suffer from degenerative loss of skeletal muscle (cachexia). Muscle wasting and loss of muscle function/performance (sarcopenia) can also occur during advanced aging. Assessing skeletal muscle mass in sarcopenia and cachexia is therefore of clinical interest for risk stratification. In comparison with fat, body fluids and bone, quantifying the skeletal muscle mass is more challenging. Computed tomography (CT) is one of the gold standard techniques for cancer diagnostics and analysis of progression, and therefore a valuable source of imaging for in vivo quantification of skeletal muscle mass. In this paper, we design a novel deep neural network-based algorithm for the automated segmentation of skeletal muscle in axial CT images at the third lumbar (L3) and the fourth thoracic (T4) levels. A two-branch network with two training steps is investigated. The network’s performance is evaluated for three trained models on separate datasets. These datasets were generated by different CT devices and data acquisition settings. To ensure the model’s robustness, each trained model was tested on all three available test sets. Errors and the effect of labeling protocol in these cases were analyzed and reported. The best performance of the proposed algorithm was achieved on 1327 L3 test samples with an overlap Jaccard score of 98% and sensitivity and specificity greater than 99%.

We present novel approaches of implementing state-of-the-art deep learning techniques for the processing of optical coherence tomography angiography (OCT-A) images for the classification of diabetic retinopathy (DR) severity. The effects of feature-engineering on a deep neural network’s classification performance is compared against unprocessed OCT-A images. We investigate the effects of lower axial resolution (simulated by using a narrower spectral bandwidth) on the classification of DR severity, and the recovery of lost features using a generative adversarial network. We also explore the relationship between DR severity classification and lateral resolution.

Highlights • FDG-PET can detect cerebral glucose hypometabolism in presymptomatic C9orf72 carriers. • Metabolic changes are seen in cingulate gyrus, frontal and temporal cortices and thalami. • Glucose metabolic changes are detectable up to 10 years prior to symptom onset. • Glucose metabolic changes are detectable prior to changes in gray matter volume., Objective Our aim is to investigate patterns of brain glucose metabolism using fluorodeoxyglucose positron emission tomography (FDG-PET) in presymptomatic carriers of the C9orf72 repeat expansion to better understand the early preclinical stages of frontotemporal dementia (FTD). Methods Structural MRI and FDG-PET were performed on clinically asymptomatic members of families with FTD caused by the C9orf72 repeat expansion (15 presymptomatic mutation carriers, C9orf72+; 20 non-carriers, C9orf72-). Regional glucose metabolism in cerebral and cerebellar gray matter was compared between groups. Results The mean age of the C9orf72+ and C9orf72- groups were 45.3 ± 10.6 and 56.0 ± 11.0 years respectively, and the mean age of FTD onset in their families was 56 ± 7 years. Compared to non-carrier controls, the C9orf72+ group exhibited regional hypometabolism, primarily involving the cingulate gyrus, frontal and temporal neocortices (left > right) and bilateral thalami. Conclusions The C9orf72 repeat expansion is associated with changes in brain glucose metabolism that are demonstrable up to 10 years prior to symptom onset and before changes in gray matter volume become significant. These findings indicate that FDG-PET may be a particularly sensitive and useful method for investigating and monitoring the earliest stages of FTD in individuals with this underlying genetic basis.

Spine disease is a growing problem in modern society and has been debilitating for every age-group. Researches have shown that more than 266 million people are facing degenerative spine disease and low back pain. CT scanning is a fast, painless, non-invasive diagnostic imaging modality that provides high spatial accuracy in obtaining the 3D structure of the vertebral. However, in real-life scenario, the clinic CT image might not cover the whole spine and the field of view might be hard to determine. Henceforth, this project aims to create and validate an automatic method that can detect, locate, and classify each vertebra from the partial field of view using deep learning. We used Mask R-CNN, a deep neural network aimed to solve the instance segmentation problem in machine learning or computer vision, and produce features such as bounding boxes, classes, and masks to identify each vertebra. This auto-detection method was validated on an open source dataset which has been used on Computational Spine Imaging (CSI 2014). The dataset was chosen by a radiologist with an eight years involvement with thoracic and lumbar spine column scans, and the data of twenty patients were collected using standard CT scanning protocol. The accuracy of the vertebra mask on 210 test images has been increased up to 99.9% DICE Coefficient in Mask R-CNN compare with 69.2% Dice Coefficient in another Deep-learning-based semantic segmentation framework U-Net.

We investigated differences due to sex in brain structural volume and cortical thickness in older cognitively normal (N=742), cognitively impaired (MCI; N=540) and Alzheimer's Dementia (AD; N=402) individuals from the ADNI and AIBL datasets (861 Males and 823 Females). General linear models were used to control the effect of relevant covariates including age, intracranial volume, magnetic resonance imaging (MRI) scanner field strength and scanner types. Significant volumetric differences due to sex were observed within different cortical and subcortical regions of the cognitively normal group. The number of significantly different regions was reduced in the MCI group, and no region remained different in the AD group. Cortical thickness was overall thinner in males than females in the cognitively normal group, and likewise, the differences due to sex were reduced in the MCI and AD groups. These findings were sustained after including cerebrospinal fluid (CSF) Tau and phosphorylated tau (pTau) as additional covariates.

Methods: Alzheimer’s disease (AD) is a worldwide prevalent age-related neurodegenerative disease with no available cure yet. Early prognosis is therefore crucial for planning proper clinical intervention. It is especially true for people diagnosed with mild cognitive impairment, to whom the prediction of whether and when the future disease onset would happen is particularly valuable. However, such prognostic prediction has been proven to be challenging, and previous studies have only achieved limited success. Approach: In this study, we seek to extract the principal component of the longitudinal disease progression trajectory in the early stage of AD, measured as the magnetic resonance imaging (MRI)-derived structural volume, to predict the onset of AD for mild cognitive impaired patients two years ahead. Results: Cross-validation results of LASSO regression using the longitudinal functional principal component (FPC) features show significant improved predictive power compared to training using the baseline volume 12 months before AD conversion [area under the receiver operating characteristic curve (AUC) of 0.802 versus 0.732] and 24 months before AD conversion (AUC of 0.816 versus 0.717). Conclusions: We present a framework using the FPCA to extract features from MRI-derived information collected from multiple timepoints. The results of our study demonstrate the advantageous predictive power of the population-based longitudinal features to predict the disease onset compared with using only cross-sectional data-based on volumetric features extracted from a single timepoint, demonstrating the improved prediction power using FPC-derived longitudinal features.

Alzheimer's disease (AD) is the most prevalent form of dementia, accounting for 60-70% of all dementias. AD is often under-diagnosed and recognized only at a later, more advanced stage, and this delay in diagnosis has been suggested as a contributing factor in the numerous unsuccessful AD treatment trials. Although there is no known cure for AD, early diagnosis is important for disease management and care. A hallmark of AD is the deposition of amyloid-β (Aβ)-containing senile neuritic plaques and neurofibrillary tangles composed of hyperphosporylated tau in the brain. However, current in vivo methods to quantify Aβ in the brain are invasive, requiring radioactive tracers and positron emission tomography. Toward development of alternative methods to assess AD progression, we focus on the retinal manifestation of AD pathology. The retina is an extension of the central nervous system uniquely accessible to light-based, non-invasive ophthalmic imaging. However, earlier studies in human retina indicate that the literature is divided on the presence of Aβ in the AD retina. To help resolve this disparity, this study assessed retinal tissues from neuropathologically confirmed AD cases to determine the regional distribution of Aβ in retinal wholemounts and to inform on future retinal image studies targeting Aβ. Concurrent post-mortem brain tissues were also collected. Neuropathological cortical assessments including neuritic plaque (NP) scores and cerebral amyloid angiopathy (CAA) were correlated with retinal Aβ using immunohistochemistry, confocal microscopy, and quantitative image analysis. Aβ load was compared between AD and control (non-AD) eyes. Our results indicate that levels of intracellular and extracellular Aβ retinal deposits were significantly higher in AD than controls. Mid-peripheral Aβ levels were greater than central retina in both AD and control eyes. In AD retina, higher intracellular Aβ was associated with lower NP score, while higher extracellular Aβ was associated with higher CAA score. Our data support the feasibility of using the retinal tissue to assess ocular Aβ as a surrogate measure of Aβ in the brain of individuals with AD. Specifically, mid-peripheral retina possesses more Aβ deposition than central retina, and thus may be the optimal location for future in vivo ocular imaging.

Optical coherence tomography (OCT) has become an essential tool in the evaluation of glaucoma, typically through analyzing retinal nerve fiber layer changes in circumpapillary scans. Three-dimensional OCT volumes enable a much more thorough analysis of the optic nerve head (ONH) region, which may be the site of initial glaucomatous optic nerve damage. Automated analysis of this region is of great interest, though large anatomical variations and the termination of layers make the requisite peripapillary layer and Bruch’s membrane opening (BMO) segmentation a challenging task. Several machine learning-based segmentation methods have been proposed for retinal layer segmentation, and a few for the ONH region, but they typically depend on either heavily averaged or pre-processed B-scans or a large amount of annotated data, which is a tedious task and resource-intensive. We evaluated a semi-supervised adversarial deep learning method for segmenting peripapillary retinal layers in OCT B-scans to take advantage of unlabeled data. We show that the use of a generative adversarial network and unlabeled data can improve the performance of segmentation. Additionally, we use a Faster R-CNN architecture to automatically segment the BMO. The proposed methods are then used for the 3D morphometric analysis of both control and glaucomatous ONH volumes to demonstrate the potential for clinical utility.

Methods:Alzheimer's disease and Frontotemporal dementia are the first and third most common forms of dementia. Due to their similar clinical symptoms, they are easily misdiagnosed as each other even with sophisticated clinical guidelines. For disease-specific intervention and treatment, it is essential to develop a computer-aided system to improve the accuracy of their differential diagnosis. Recent advances in deep learning have delivered some of the best performance for medical image recognition tasks. However, its application to the differential diagnosis of AD and FTD pathology has not been explored. Approach:In this study, we proposed a novel deep learning based framework to distinguish between brain images of normal aging individuals and subjects with AD and FTD. Specifically, we combined the multi-scale and multi-type MRI-base image features with Generative Adversarial Network data augmentation technique to improve the differential diagnosis accuracy. Results:Each of the multi-scale, multitype, and data augmentation methods improved the ability for differential diagnosis for both AD and FTD. A 10-fold cross validation experiment performed on a large sample of 1,954 images using the proposed framework achieved a high overall accuracy of 88.28%. Conclusions:The salient contributions of this study are three-fold: (1) our experiments demonstrate that the combination of multiple structural features extracted at different scales with our proposed deep neural network yields superior performance than individual features; (2) we show that the use of Generative Adversarial Network for data augmentation could further improve the discriminant ability of the network regarding challenging tasks such as differentiating dementia sub-types; (3) and finally, we show that ensemble classifier strategy could make the network more robust and stable.

We present updates upon our novel machine-learning methods for the acquisition, processing, and classification of Optical Coherence Tomography Angiography (OCT-A) images. Transitioning from traditional registration methods to machine-learning based methods provided significant reductions in computation time for serial image acquisition and averaging. Through a vessel segmentation network, clinically useful parameters were extracted and then fed to our classification network which was able to classify different diabetic retinopathy severities. The DNN pipeline was also implemented on data acquired with Sensorless Adaptive Optics OCT-A. This work has potential to subsequently reduce clinical overhead and help expedite treatments, resulting in improved patient prognoses.

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.

Alzheimer’s disease (AD) is one of the most common neurodegenerative diseases with a commonly seen prodromal mild cognitive impairment (MCI) phase where memory loss is the main complaint progressively worsening with behavior issues and poor self-care. However, not all individuals clinically diagnosed with MCI progress to AD. A fraction of subjects with MCI either progress to non-AD dementia or remain stable at the MCI stage without progressing to dementia. Although a curative treatment of AD is currently unavailable, it is extremely important to correctly identify the individuals in the MCI phase that will go on to develop AD so that they may benefit from a curative treatment when one becomes available in the near future. At the same time, it would be highly desirable to also correctly identify those in the MCI phase that do not have AD pathology so they may be spared from unnecessary pharmocologic interventions that, at best, may provide them no benefit, and at worse, could further harm them with adverse side-effects. Additionally, it may be easier and simpler to identify the cause of the cognitive impairment in these non-AD cases, and hence proper identification of prodromal AD will be of benefit to these individuals as well. Fluorodeoxy glucose positron emission tomography (FDG-PET) captures the metabolic activity of the brain, and this imaging modality has been reported to identify changes related to AD prior to the onset of structural changes. Prior work on designing classifier using FDG-PET imaging has been promising. Since deep-learning has recently emerged as a powerful tool to mine features and use them for accurate labeling of the group membership of given images, we propose a novel deep-learning framework using FDG-PET metabolism imaging to identify subjects at the MCI stage with presymptomatic AD and discriminate them from other subjects with MCI (non-AD / non-progressive). Our multiscale deep neural network obtained 82.51% accuracy of classification just using measures from a single modality (FDG-PET metabolism data) outperforming other comparable FDG-PET classifiers published in the recent literature.

Fluorodeoxyglucose positron emission tomography (FDG-PET) imaging based 3D topographic brain glucose metabolism patterns from normal controls (NC) and individuals with dementia of Alzheimer's type (DAT) are used to train a novel multi-scale ensemble classification model. This ensemble model outputs a FDG-PET DAT score (FPDS) between 0 and 1 denoting the probability of a subject to be clinically diagnosed with DAT based on their metabolism profile. A novel 7 group image stratification scheme is devised that groups images not only based on their associated clinical diagnosis but also on past and future trajectories of the clinical diagnoses, yielding a more continuous representation of the different stages of DAT spectrum that mimics a real-world clinical setting. The potential for using FPDS as a DAT biomarker was validated on a large number of FDG-PET images (N=2984) obtained from the Alzheimer's Disease Neuroimaging Initiative (ADNI) database taken across the proposed stratification, and a good classification AUC (area under the curve) of 0.78 was achieved in distinguishing between images belonging to subjects on a DAT trajectory and those images taken from subjects not progressing to a DAT diagnosis. Further, the FPDS biomarker achieved state-of-the-art performance on the mild cognitive impairment (MCI) to DAT conversion prediction task with an AUC of 0.81, 0.80, 0.77 for the 2, 3, 5 years to conversion windows respectively., Highlights • A novel FDG-PET biomarker indicating probability of dementia of Alzheimer's type (DAT) diagnosis • A new 7 group image stratification approach representing different stages along the DAT spectrum • Comprehensive biomarker evaluation using 2984 ADNI images achieved a good validation AUC of 0.78. • Achieved an AUC of 0.81, 0.80, 0.77 for 2, 3, 5 years to conversion on MCI to DAT conversion prediction

Purpose To evaluate the performance of a federated learning framework for deep neural network-based retinal microvasculature segmentation and referable diabetic retinopathy (RDR) classification using optical coherence tomography (OCT) and OCT angiography (OCTA). Design Retrospective analysis of clinical OCT and OCTA scans of control and diabetic patients. Participants The data used for microvasculature segmentation consisted of OCTA en face images acquired from four OCT instruments with a total of 153 images, with fields-of-view ranging from 2×2mm to 6×6mm. The data used for RDR classification consisted of OCTA en face images and structural OCT projections acquired from two commercial OCT systems with a total of 700 images. Methods OCTA images used for microvasculature segmentation were manually delineated and verified by retina experts. Diabetic retinopathy (DR) severity was evaluated by retinal specialists and condensed into two classes: non-referable DR, and RDR. The federated learning configuration was demonstrated via simulation using four clients for microvasculature segmentation and compared with other collaborative methods. Subsequently, federated learning was applied over multiple institutions for RDR classification, and was compared with models trained and tested on data from the same institution (internal models) and different institutions (external models). Main Outcome Measures:For microvasculature segmentation, we measured the accuracy and Dice similarity coefficient (DSC). For severity classification, we measured accuracy, area under the receiver operating characteristic (AUROC) curve, area under the precision-recall curve, balanced accuracy, F1 score, sensitivity, specificity, and accuracy of further stratified severities. Results For both applications, federated learning achieved similar performance to internal models. Additionally, for microvasculature segmentation, the federated learning model achieved similar performance (mean DSC across all test sets 0.793) to models trained on a fully-centralized dataset (mean DSC: 0.807). For RDR classification, federated learning achieved an AUROC of 0.944–0.970; the internal models attained an AUROC of 0.945-0.978. Similar results are reflected in the other calculated evaluation metrics. Conclusions Federated learning shows similar results to traditional deep learning in both applications of segmentation and classification, while maintaining data privacy. Evaluation metrics highlight the potential of collaborative learning for increasing domain diversity and the generalizability of models used for the classification of OCT data.

Computer-assistant diagnosis of retinal disease relies heavily on the accurate detection of retinal boundaries and other pathological features such as fluid accumulation. Optical coherence tomography (OCT) is a non-invasive ophthalmological imaging technique that has become a standard modality in the field due to its ability to detect cross-sectional retinal pathologies at the micrometer level. In this work, we presented a novel framework to achieve simultaneous retinal layers and fluid segmentation. A dual-branch deep neural network, termed LF-UNet, was proposed which combines the expansion path of the U-Net and original fully convolutional network, with a dilated network. In addition, we introduced a cascaded network framework to include the anatomical awareness embedded in the volumetric image. Cross validation experiments showed that the proposed LF-UNet has superior performance compared to the state-of-the-art methods, and that incorporating the relative positional map structural prior information could further improve the performance regardless of the network. The generalizability of the proposed network was demonstrated on an independent dataset acquired from the same types of device with different field of view, or images acquired from different device.

PURPOSE To demonstrate the utility of swept-source anterior segment optical coherence tomography (SS-AS-OCT) in guiding placement of transconjunctival sutures in hypotonous patients after a trabeculectomy. PATIENTS AND METHODS This is a longitudinal case series of 10 eyes from 10 patients who required transconjunctival sutures after a trabeculectomy. SS-AS-OCT was used to aid in the placement of the sutures to improve the function of the overfiltering bleb. RESULTS SS-AS-OCT reliably identified localized areas of overfiltering, allowing for targeted suture placement in 8 eyes. The 2 eyes in which localized areas of overfiltering were not found required further surgical intervention. CONCLUSIONS SS-AS-OCT enhances transconjunctival suturing for overfiltering blebs when focal fluid accumulation is visualized.

International audience; We propose a novel approach for quantitative shape variability analysis in reti-nal optical coherence tomography images using the functional shape (fshape) framework. The fshape framework uses surface geometry together with functional measures, such as retinal layer thickness defined on the layer surface, for registration across anatomical shapes. This is used to generate a population mean template of the geometry-function measures from each individual. Shape variability across multiple retinas can be measured by the geometrical deformation and functional residual between the template and each of the observations. To demonstrate the clinical relevance and application of the framework, we generated atlases of the inner layer surface and layer thickness of the Retinal Nerve Fiber Layer (RNFL) of glaucomatous and normal subjects, visualizing detailed spatial pattern of RNFL loss in glaucoma. Additionally, a regularized linear discriminant analysis classifier was used to automatically classify glau-coma, glaucoma-suspect, and control cases based on RNFL fshape metrics.

Biomarkers for dementia of Alzheimer's type (DAT) are sought to facilitate accurate prediction of the disease onset, ideally predating the onset of cognitive deterioration. T1‐weighted magnetic resonance imaging (MRI) is a commonly used neuroimaging modality for measuring brain structure in vivo, potentially providing information enabling the design of biomarkers for DAT. We propose a novel biomarker using structural MRI volume‐based features to compute a similarity score for the individual's structural patterns relative to those observed in the DAT group. We employed ensemble‐learning framework that combines structural features in most discriminative ROIs to create an aggregate measure of neurodegeneration in the brain. This classifier is trained on 423 stable normal control (NC) and 330 DAT subjects, where clinical diagnosis is likely to have the highest certainty. Independent validation on 8,834 unseen images from ADNI, AIBL, OASIS, and MIRIAD Alzheimer's disease (AD) databases showed promising potential to predict the development of DAT depending on the time‐to‐conversion (TTC). Classification performance on stable versus progressive mild cognitive impairment (MCI) groups achieved an AUC of 0.81 for TTC of 6 months and 0.73 for TTC of up to 7 years, achieving state‐of‐the‐art results. The output score, indicating similarity to patterns seen in DAT, provides an intuitive measure of how closely the individual's brain features resemble the DAT group. This score can be used for assessing the presence of AD structural atrophy patterns in normal aging and MCI stages, as well as monitoring the progression of the individual's brain along with the disease course.

Purpose To evaluate the role of ensemble learning techniques with deep learning in classifying diabetic retinopathy (DR) in optical coherence tomography angiography (OCTA) images and their corresponding co-registered structural images. Methods A total of 463 volumes from 380 eyes were acquired using the 3 × 3-mm OCTA protocol on the Zeiss Plex Elite system. Enface images of the superficial and deep capillary plexus were exported from both the optical coherence tomography and OCTA data. Component neural networks were constructed using single data-types and fine-tuned using VGG19, ResNet50, and DenseNet architectures pretrained on ImageNet weights. These networks were then ensembled using majority soft voting and stacking techniques. Results were compared with a classifier using manually engineered features. Class activation maps (CAMs) were created using the original CAM algorithm and Grad-CAM. Results The networks trained with the VGG19 architecture outperformed the networks trained on deeper architectures. Ensemble networks constructed using the four fine-tuned VGG19 architectures achieved accuracies of 0.92 and 0.90 for the majority soft voting and stacking methods respectively. Both ensemble methods outperformed the highest single data-type network and the network trained on hand-crafted features. Grad-CAM was shown to more accurately highlight areas of disease. Conclusions Ensemble learning increases the predictive accuracy of CNNs for classifying referable DR on OCTA datasets. Translational relevance Because the diagnostic accuracy of OCTA images is shown to be greater than the manually extracted features currently used in the literature, the proposed methods may be beneficial toward developing clinically valuable solutions for DR diagnoses.

Background Body composition from computed tomography (CT) scans is associated with cancer outcomes including surgical complications, chemotoxicity, and survival. Most studies manually segment CT scans, but Automatic Body composition Analyser using Computed tomography image Segmentation (ABACS) software automatically segments muscle and adipose tissues to speed analysis. Here, we externally evaluate ABACS in an independent dataset. Methods Among patients with non‐metastatic colorectal (n = 3102) and breast (n = 2888) cancer diagnosed from 2005 to 2013 at Kaiser Permanente, expert raters annotated tissue areas at the third lumbar vertebra (L3). To compare ABACS segmentation results to manual analysis, we quantified the proportion of pixel‐level image overlap using Jaccard scores and agreement between methods using intra‐class correlation coefficients for continuous tissue areas. We examined performance overall and among subgroups defined by patient and imaging characteristics. To compare the strength of the mortality associations obtained from ABACS's segmentations to manual analysis, we computed Cox proportional hazards ratios (HRs) and 95% confidence intervals (95% CI) by tertile of tissue area. Results Mean ± SD age was 63 ± 11 years for colorectal cancer patients and 56 ± 12 for breast cancer patients. There was strong agreement between manual and automatic segmentations overall and within subgroups of age, sex, body mass index, and cancer stage: average Jaccard scores and intra‐class correlation coefficients exceeded 90% for all tissues. ABACS underestimated muscle and visceral and subcutaneous adipose tissue areas by 1–2% versus manual analysis: mean differences were small at −2.35, −1.97 and −2.38 cm2, respectively. ABACS's performance was lowest for the

High quality visualization of the retinal microvasculature can improve our understanding of the onset and development of retinal vascular diseases, especially Diabetic Retinopathy (DR), which is a major cause of visual morbidity and is increasing in prevalence. Optical Coherence Tomography Angiography (OCT-A) images are acquired over multiple seconds and are particularly susceptible to motion artifacts, which are more prevalent when imaging individuals with DR whose ability to fixate is limited due to deteriorating vision. The sequential acquisition and averaging of multiple OCT-A images can be performed for removing motion artifact and increasing the contrast of the vascular network. As motion artifacts often irreversibly corrupt OCT-A images of DR eyes, a robust registration pipeline is needed before feature preserving image averaging can be performed. In this report we present an improvement upon a novel method for the acquisition, processing, segmentation, registration, and averaging of sequentially acquired OCT-A images, to correct for motion artifacts in images of DR eyes. Image discontinuities caused by rapid micro-saccadic movements and image warping due to smoother reflex movements were corrected by strip-wise affine registration and subsequent local similarity-based non-rigid registration. Where our previous work was limited by the need for at least one image containing no motion artifact, thus reducing its clinical relevance, this novel template-less method stitches together partial images to form complete, motion-free images. These techniques significantly improve image quality, increasing the value for clinical diagnosis and increasing the range of patients for whom high quality OCT-A images can be acquired.

We report on the ongoing project "PREDICT-ADFTD: Multimodal Imaging Prediction of AD/FTD and Differential Diagnosis" describing completed and future work supported by this grant. This project is a multi-site, multi-study collaboration effort with research spanning seven sites across the US and Canada. The overall goal of the project is to study neurodegeneration within Alzheimer's Disease, Frontotemporal Dementia, and related neurodegenerative disorders, using a variety of brain imaging and computational techniques to develop methods for the early and accurate prediction of disease and its course. The overarching goal of the project is to develop the earliest and most accurate biomarker that can differentiate clinical diagnoses to inform clinical trials and patient care. In its third year, this project has already completed several projects to achieve this goal, focusing on (1) structural MRI (2) machine learning and (3) FDG-PET and multimodal imaging. Studies utilizing structural MRI have identified key features of underlying pathology by studying hippocampal deformation that is unique to clinical diagnosis and also post-mortem confirmed neuropathology. Several machine learning experiments have shown high classification accuracy in the prediction of disease based on Convolutional Neural Networks utilizing MRI images as input. In addition, we have also achieved high accuracy in predicting conversion to DAT up to five years in the future. Further, we evaluated multimodal models that combine structural and FDG-PET imaging, in order to compare the predictive power of multimodal to unimodal models. Studies utilizing FDG-PET have shown significant predictive ability in the prediction and progression of disease.

Artificial intelligence (AI) has a great impact on our daily living and makes our lives more efficient and productive. Especially during the coronavirus disease (COVID-19) pandemic, AI has played a key role in response to the global health crisis. There has been a boom in AI innovation and its use since the pandemic. However, despite its widespread adoption and great potential, most people have little knowledge of AI concepts and realization of its potential. The objective of this white paper is to communicate the importance of AI and its benefits to society. The report covers AI applications in six different topics from medicine (AI deployment in clinical settings, imaging and diagnostics, and acceleration of drug discovery) to more social aspects (support older adults in long-term care homes, and AI in supporting small and medium enterprises. The report ends with nine steps to consider for moving forward with AI implementation during and post pandemic period. These include legal and ethical data collection and storage, greater data access, multidisciplinary collaboration, and policy reform.

To evaluate the quantitative characteristics of the radial peripapillary capillary (RPC) network in glaucoma, glaucoma suspect, and normal eyes using speckle variance optical coherence tomography angiography (OCT-A). To determine correlations between RPC density, nerve fiber layer (NFL) thickness, and visual field indices.Cross-sectional study.OCT-A images of RPCs were acquired at a single institution using a custom-built 1060 nm system from 3 groups: unilateral glaucoma (10 eyes from 5 subjects), glaucoma suspects (6 eyes from 3 subjects), and normal control eyes (16 eyes from 9 normal subjects). Peripapillary NFL thickness measurements were determined using spectral-domain optical coherence tomography. Glaucoma and glaucoma suspects also underwent automated 30-2 Humphrey visual field analysis. Manual tracing techniques were used to quantify RPC density in the OCT-A images. Data were analyzed using a linear mixed model with 1 fixed-effect covariate. Correlations between main outcome measures (RPC density, NFL thickness, and visual field index) were determined.Mean age was not significantly different between the 3 groups (P = .25). The density of RPCs was significantly lower in glaucomatous eyes compared with matched-peripapillary regions in the fellow eye, glaucoma suspect group, and normal group (all P.001). RPC density was strongly correlated with NFL thickness (P.001) and visual field index (P.001).Significant reductions in RPC density were correlated with sites of NFL decrease and visual field loss in glaucoma. Speckle variance OCT-A allows visualization and quantification of RPCs and may therefore be a useful tool for indirectly quantifying and monitoring retinal ganglion cell axonal injury in glaucoma.

Automated measurements of the human cone mosaic requires the identification of individual cone photoreceptors. The current gold standard, manual labeling, is a tedious process and can not be done in a clinically useful timeframe. As such, we present an automated algorithm for identifying cone photoreceptors in adaptive optics optical coherence tomography (AO-OCT) images. Our approach fine-tunes a pre-trained convolutional neural network originally trained on AO scanning laser ophthalmoscope (AO-SLO) images, to work on previously unseen data from a different imaging modality. On average, the automated method correctly identified 94% of manually labeled cones when compared to manual raters, from twenty different AO-OCT images acquired from five normal subjects. Voronoi analysis confirmed the general hexagonal-packing structure of the cone mosaic as well as the general cone density variability across portions of the retina. The consistency of our measurements demonstrates the high reliability and practical utility of having an automated solution to this problem.

As a non-invasive imaging modality, optical coherence tomography (OCT) can provide micrometer-resolution 3D images of retinal structures. These images can help reveal disease-related alterations below the surface of the retina, such as the presence of edema, or accumulation of fluid which can distort vision, and are an indication of disruptions in the vasculature of the retina. In this paper, a new framework is proposed for multiclass fluid segmentation and detection in the retinal OCT images. Based on the intensity of OCT images and retinal layer segmentations provided by a graph-cut algorithm, a fully convolutional neural network was trained to recognize and label the fluid pixels. Random forest classification was performed on the segmented fluid regions to detect and reject the falsely labeled fluid regions. The proposed framework won the first place in the MICCAI RETOUCH challenge in 2017 on both the segmentation performance (mean Dice: 0.7667) and the detection performance (mean AUC: 1.00) tasks.

Frontotemporal dementia (FTD) is a neurodegenerative disease with a strong genetic basis. Understanding the structural brain changes during pre-symptomatic stages may allow for earlier diagnosis of patients suffering from FTD; therefore, we investigated asymptomatic members of FTD families with mutations in C9orf72 and granulin (GRN) genes. Clinically asymptomatic subjects from families with C9orf72 mutation (15 mutation carriers, C9orf72+; and 23 non-carriers, C9orf72−) and GRN mutations (9 mutation carriers, GRN+; and 15 non-carriers, GRN−) underwent structural neuroimaging (MRI). Cortical thickness and subcortical gray matter volumes were calculated using FreeSurfer. Group differences were evaluated, correcting for age, sex and years to mean age of disease onset within the subject's family. Mean age of C9orf72+ and C9orf72− were 42.6 ± 11.3 and 49.7 ± 15.5 years, respectively; while GRN+ and GRN− groups were 50.1 ± 8.7 and 53.2 ± 11.2 years respectively. The C9orf72+ group exhibited cortical thinning in the temporal, parietal and frontal regions, as well as reduced volumes of bilateral thalamus and left caudate compared to the entire group of mutation non-carriers (NC: C9orf72− and GRN− combined). In contrast, the GRN+ group did not show any significant differences compared to NC. C9orf72 mutation carriers demonstrate a pattern of reduced gray matter on MRI prior to symptom onset compared to GRN mutation carriers. These findings suggest that the preclinical course of FTD differs depending on the genetic basis and that the choice of neuroimaging biomarkers for FTD may need to take into account the specific genes involved in causing the disease., Highlights • Patterns of brain atrophy differ in GRN & C9orf72 presymptomatic carriers. • Asymptomatic C9orf72+ carriers exhibit gray matter atrophy at a relatively early age. • GRN+ carrier did not demonstrate any significant gray matter changes despite being relatively older. • Preclinical course of FTD & the choice of imaging biomarker differ based on genetic mutations.

Nonrigid registration of optical coherence tomography (OCT) images is an important problem in studying eye diseases, evaluating the effect of pharmaceuticals in treating vision loss, and performing group-wise cross-sectional analysis. High dimensional nonrigid registration algorithms required for cross-sectional and longitudinal analysis are still being developed for accurate registration of OCT image volumes, with the speckle noise in images presenting a challenge for registration. Development of algorithms for segmentation of OCT images to generate surface models of retinal layers has advanced considerably and several algorithms are now available that can segment retinal OCT images into constituent retinal surfaces. Important morphometric measurements can be extracted if accurate surface registration algorithm for registering retinal surfaces onto corresponding template surfaces were available. In this paper, we present a novel method to perform multiple and simultaneous retinal surface registration, targeted to registering surfaces extracted from ocular volumetric OCT images. This enables a point-to-point correspondence (homology) between template and subject surfaces, allowing for a direct, vertex-wise comparison of morphometric measurements across subject groups. We demonstrate that this approach can be used to localize and analyze regional changes in choroidal and nerve fiber layer thickness among healthy and glaucomatous subjects, allowing for cross-sectional population wise analysis. We also demonstrate the method's ability to track longitudinal changes in optic nerve head morphometry, allowing for within-individual tracking of morphometric changes. This method can also, in the future, be used as a precursor to 3-D OCT image registration to better initialize nonrigid image registration algorithms closer to the desired solution.

Voxel-based morphometry (VBM) and tensor-based morphometry (TBM) both rely on spatial normalization to a template and yet have different requirements for the level of registration accuracy. VBM requires only global alignment of brain structures, with limited degrees of freedom in transformation, whereas TBM performs best when the registration is highly deformable and can achieve higher registration accuracy. In addition, the registration accuracy varies over the whole brain, with higher accuracy typically observed in subcortical areas and lower accuracy seen in cortical areas. Hence, even the determinant of Jacobian of registration maps is spatially varying in their accuracy, and combining these with VBM by direct multiplication introduces errors in VBM maps where the registration is inaccurate. We propose a unified approach to combining these 2 morphometry methods that is motivated by these differing requirements for registration and our interest in harnessing the advantages of both. Our novel method uses local estimates of registration confidence to determine how to weight the influence of VBM- and TBM-like approaches. Results are shown on healthy and mild Alzheimer's subjects (N = 150) investigating age and group differences, and potential of differential diagnosis is shown on a set of Alzheimer's disease (N = 34) and frontotemporal dementia (N = 30) patients compared against controls (N = 14). These show that the group differences detected by our proposed approach are more descriptive than those detected from VBM, Jacobian-modulated VBM, and TBM separately, hence leveraging the advantages of both approaches in a unified framework.

Both genetic variants and brain region abnormalities are recognized to play a role in cognitive decline. We explore the association between singlenucleotide polymorphisms (SNPs) in linkage regions for Alzheimer’s disease and rates of decline in brain structure using data from the Alzheimers Disease Neuroimaging Initiative (ADNI).In an initial discovery stage, we assessed the presence of linear association between the minor allele counts of 75,845 SNPs in the Alzgene linkage regions and predicted rates of change in structural MRI measurements for 56 brain regions using an RV test. In a second, refinement stage, we reduced the number of SNPs using a bootstrap-enhanced sparse canonical correlation analysis (SCCA) with a fixed tuning parameter. Each SNP was assigned an importance measure proportional to the number of times it was estimated to have a nonzero coefficient in repeated re-sampling from the ADNI-1 sample. We created refined lists of SNPs based on importance probabilities greater than 50% and 90%, respectively. In a third, validation stage, we assessed the multivariate association between these refined lists of SNPs and the rates of structural change in the independent ADNI-2 study dataset.There was strong statistical evidence for linear association between the SNPs in the Alzgene linkage regions and the 56 imaging phenotypes in both the ADNI-1 and ADNI-2 samples (p < 0.0001). The bootstrap-enhanced SCCA identified 1,694 priority SNPs with importance probabilities > 50% and 22 SNPs with importance probabilities > 90%. The 1,694 prioritized SNPs in the ADNI-1 data were associated with imaging phenotypes in the ADNI-2 data (p = 0.0021).This manuscript presents an analysis that addresses challenges in current imaging genetics studies such as biased sampling designs and highdimensional data with low-signal. Genes corresponding to priority SNPs having the highest contribution in the validation data have previously been implicated or hypothesized to be implicated in AD, including GCLC, IDE, and STAMBP1andFAS. We hypothesize that the effect sizes of the 1,694 SNPs in the priority set are likely small, but further investigation within this set may advance understanding of the missing heritability in late-onset Alzheimers disease. Multivariate analysis; Linkage regions; Imaging genetics; Endophenotypes; Inverse probability weighting; Variable importance probabilities

Background: Recent work on Alzheimer's disease (AD) diagnosis focuses on neuroimaging modalities; however, these methods are expensive, invasive, and not available to all patients. Ocular imaging of biomarkers, such as drusen in the peripheral retina, could provide an alternative method to diagnose AD. Objective: This study compares macular and peripheral drusen load in control and AD eyes. Methods: Postmortem eye tissues were obtained from donors with a neuropathological diagnosis of AD. Retina from normal donors were processed and categorized into younger (55 years) groups. After fixation and dissection, 3-6 mm punches of RPE/choroid were taken in macular and peripheral (temporal, superior, and inferior) retinal regions. Oil red O positive drusen were counted and grouped into two size categories: small ( Results: There was a significant increase in the total number of macular and peripheral hard drusen in older, compared to younger, normal eyes (p Conclusion: Imaging temporal drusen in the eye may have benefit for diagnosing and monitoring progression of AD. Our results on CAA severity and temporal intermediate drusen in the AD eye are novel. Future studies are needed to further understand the interactions among CAA and drusen formation.

High quality visualization of the retinal microvasculature can improve our understanding of the onset and development of retinal vascular diseases, which are a major cause of visual morbidity and are increasing in prevalence. Optical Coherence Tomography Angiography (OCT-A) images are acquired over multiple seconds and are particularly susceptible to motion artifacts, which are more prevalent when imaging patients with pathology whose ability to fixate is limited. The acquisition of multiple OCT-A images sequentially can be performed for the purpose of removing motion artifact and increasing the contrast of the vascular network through averaging. Due to the motion artifacts, a robust registration pipeline is needed before feature preserving image averaging can be performed. In this report, we present a novel method for a GPU-accelerated pipeline for acquisition, processing, segmentation, and registration of multiple, sequentially acquired OCT-A images to correct for the motion artifacts in individual images for the purpose of averaging. High performance computing, blending CPU and GPU, was introduced to accelerate processing in order to provide high quality visualization of the retinal microvasculature and to enable a more accurate quantitative analysis in a clinically useful time frame. Specifically, image discontinuities caused by rapid micro-saccadic movements and image warping due to smoother reflex movements were corrected by strip-wise affine registration estimated using Scale Invariant Feature Transform (SIFT) keypoints and subsequent local similarity-based non-rigid registration. These techniques improve the image quality, increasing the value for clinical diagnosis and increasing the range of patients for whom high quality OCT-A images can be acquired.