Data Pooling and Sampling of Heterogeneous Image Data for White Matter Hyperintensity Segmentation

Introduction: Patients with non-ischemic DCM are at risk of life-threatening ventricular arrhythmias. Improved patient selection for ICD implantation is warranted. Deep neural networks (DNN) can discover complex ECG features without the need for hand-crafted feature extraction. This study aimed to distinguish patients at risk of arrhythmias and detect and visualize ECG characteristics that attribute to the DNN classification. Methods: We trained a DNN model on baseline raw ECG samples of adult patients with DCM for the composite outcome of sustained ventricular arrhythmia, (aborted) sudden cardiac death and appropriate ICD therapy in the University Medical Centre Utrecht. Gradient-Weighted Class Activation Mapping was used for visual interpretation of the network. DNN performance was validated both internally and in an external cohort from the Maastricht University Medical University Center. Results: 26.628 ECGs were used for pre-training the network. In total, 317 patients with DCM were included in the primary cohort with a median age of 52 years (IQR 42 - 61), median LVEF of 25% (IQR 20-33), NYHA II and III (32% and 25%). 233 patients were ICD recipients. Subsequently 84 patients reached the outcome (median follow-up duration of 3.96 years (IQR 1.94 - 7.00)). The DNN focused on subtle but predictive differences in the terminal QRS-segment, the T-wave, and the P-wave (figure). The AUC of the DNN was 0.70 in cross-validation. External validation was similar, with an AUC of 0.66. Conclusion: In this study we trained and validated a deep neural network for prediction of life-threatening arrhythmic events. Even though the discriminatory changes on ECG segments were very subtle, the trained DNN was able to predict life-threatening events, focusing on the terminal QRS-segment, the T-wave, and the P-wave. Future studies are warranted to elucidate the identified ECG abnormalities and understand their electrophysiological substrates for LTA prediction in DCM.

Recently deep-learning (DL) techniques have been widely adopted in side-channel power analysis. A DL-assisted SCA generally consists of two phases: a deep neural network (DNN) training phase and a follow-on attack phase using the trained DNN. However, currently the two phases are not well aligned, as there is no conclusion on what metric used in the training can result in the most effective attack in the second phase. When traditional loss functions such as negative log-likelihood (NLL) are used in training a DNN, the trained model does not yield optimal follow-on attack. Recently some information theoretical SCA leakage metrics are proposed, either as the validation metric to stop the DNN training with traditional loss functions, or as both the validation metric and the training loss function. None of those proposed metrics, however, directly measures the SCA effectiveness. We propose to conduct DNN training directly with a common SCA effectiveness metric, Guessing Entropy (GE). We overcome the prior practical difficulty of using GE in DNN training by utilizing the GEEA estimation algorithm introduced in CHES 2020. We show that using GEEA as either the validation metric or the loss function produces DNN models that lead to much more effective follow-on attacks. Our work consolidates the DL-assisted SCA framework with a consistent metric, which shows great potential to be adopted as the universal SCA-oriented DNN training framework.

The recent surge of interest in Deep Neural Networks (DNNs) has led to increasingly complex networks that tax computational and memory resources. Many DNNs presently use 16-bit or 32-bit floating point operations. Significant performance and power gains can be obtained when DNN accelerators support low-precision numerical formats. Despite considerable research, there is still a knowledge gap on how low-precision operations can be realized for both DNN training and inference. In this work, we propose a DNN architecture, Deep Positron, with posit numerical format operating successfully at ≤8 bits for inference. We propose a precision-adaptable FPGA soft core for exact multiply-and-accumulate for uniform comparison across three numerical formats, fixed, floating-point and posit. Preliminary results demonstrate that 8-bit posit has better accuracy than 8-bit fixed or floating-point for three different low-dimensional datasets. Moreover, the accuracy is comparable to 32-bit floating-point on a Xilinx Virtex-7 FPGA device. The trade-offs between DNN performance and hardware resources, i.e. latency, power, and resource utilization, show that posit outperforms in accuracy and latency at 8-bit and below.

The sensitivity of wide parameter-space searches for continuous gravitational waves (CWs) is limited by their high computational cost. Deep learning is being studied as an alternative method to replace various aspects of a CW search. In previous work [], new design principles were presented for deep neural network (DNN) search of CWs and such DNNs were trained to perform a targeted search with matched-filtering sensitivity. In this paper, we adapt these design principles to build a DNN architecture for wide parameter-space searches in 10 days of data from two detectors (H1 and L1). We train a DNN for each of the benchmark cases: six all-sky searches and eight directed searches at different frequencies in the search band of 20–1000 Hz. We compare our results to the DNN sensitivity achieved from Dreissigacker and Prix [] and find that our trained DNNs are more sensitive in all the cases. The absolute improvement in detection probability ranges from 6.5% at 20 Hz to 38% at 1000 Hz in the all-sky cases and from 1.5% at 20 Hz to 59.4% at 500 Hz in the directed cases. An all-sky DNN trained on the entire search band of 20–1000 Hz shows a high sensitivity at all frequencies, providing a proof of concept for training a single DNN to perform the entire search. We also study the generalization of the DNN performance to signals with different signal amplitude, frequency, and the dependence of the DNN sensitivity on sky position. Published by the American Physical Society 2024

Vascular cognitive impairment is common and represents a spectrum of cognitive dysfunction associated with stroke and cardiovascular risk factors which may be slight, moderate or severe.1 Recently and as a major advance, the National Institute of Neurological Disorders and Stroke—Canadian Stroke Network Vascular Cognitive Impairment Harmonization Standards Working Group published clinical and research standards for the description and study of vascular cognitive impairment.2 In this update we report advances in vascular cognitive impairment in the following areas: clinical trials and treatment, new risk factors, white matter disease, and genetics. Cardiovascular risk factors such as blood pressure elevation and hyperglycemia are prime targets for clinical trial study because they are believed to cause negative effects on brain structure and cognitive function and may even influence risk of Alzheimer disease (AD).3 Results of recently published observational epidemiological studies have fueled controversy in relation to the role of blood pressure on cognition because some studies such as the Honolulu Asia Aging Study4 and Cache County Study5 suggest that antihypertensive therapy may reduce risk of dementia and cognitive decline or reduce incidence of AD, respectively, whereas the Religious Orders Study6 did not find an association between blood pressure and risk of AD or cognitive decline. Meta-analysis of patients with cardiovascular and/or cerebrovascular disease who received blood pressure-lowering treatment show a trend toward prevention of dementia and/or cognitive decline;7 however, a systematic analysis of 3 studies comprising 12 091 patients with hypertension who were treated with either medication or lifestyle strategies for at least 6 months show no evidence that blood pressure-lowering prevents dementia or cognitive impairment.8 Lack of definitive results in some of these trials may be explained by insufficient power to detect treatment effects, measurement error in cognitive end points, variation in treatment effects between …

Survey on Deep Neural Networks in Speech and Vision Systems

Deep Learning (DL) has become a prominent machine learning technique due to the availability of efficient computational resources in the form of Graphics Processing Units (GPUs), large-scale datasets and a variety of models. The newer generation of GPUs are being designed with special emphasis on optimizing performance for DL applications. Also, the availability of easy-to-use DL frameworks—like PyTorch and TensorFlow— has enhanced productivity of domain experts to work on their custom DL applications from diverse domains. However, existing Deep Neural Network (DNN) training approaches may not fully utilize the newly emerging powerful GPUs like the NVIDIA A100—this is the primary issue that we address in this paper. Our motivating analyses show that the GPU utilization on NVIDIA A100 can be as low as 43% using traditional DNN training approaches for small-to-medium DL models and input data size. This paper proposes AccDP—a data-parallel distributed DNN training approach—to accelerate GPU-based DL applications. AccDP exploits the Message Passing Interface (MPI) communication library coupled with the NVIDIA’s Multi-Process Service (MPS) to increase the amount of work assigned to parallel GPUs resulting in higher utilization of compute resources. We evaluate our proposed design on different small-to-medium DL models and input sizes on the state-of-the-art HPC clusters. By injecting more parallelism into DNN training using our approach, the evaluation shows up to 58% improvement in training performance on a single GPU and up to 62% on 16 GPUs compared to regular DNN training. Furthermore, we conduct an in-depth characterization to determine the impact of several DNN training factors and best practices—including the batch size and the number of data loading workers— to optimally utilize GPU devices. To the best of our knowledge, this is the first work that explores the use of MPS and MPI to maximize the utilization of GPUs in distributed DNN training.

Phase unwrapping is an important but challenging issue in phase measurement. Even with the research efforts of a few decades, unfortunately, the problem remains not well solved, especially when heavy noise and aliasing (undersampling) are present. We propose a database generation method for phase-type objects and a one-step deep learning phase unwrapping method. With a trained deep neural network, the unseen phase fields of living mouse osteoblasts and dynamic candle flame are successfully unwrapped, demonstrating that the complicated nonlinear phase unwrapping task can be directly fulfilled in one step by a single deep neural network. Excellent anti-noise and anti-aliasing performances outperforming classical methods are highlighted in this paper.

Introduction.Cerebral microangiopathy (CMA), being the leading cause of vascular cognitive impairment and strokes, has a number of causes, among which chronic kidney disease (CKD) and programmed hemodialysis (HD) are the least studied.Purpose of the study:to determine the frequency of CMA neuroimaging markers and risk factors for its development in patients receiving renal replacement therapy for a long time using the programmed HD.Material and methods:the study involved 70 patients who had been on programmed HD for 10 months or more. Clinical neurological examination, laboratory tests and brain MRI were performed. The analysis of CMA neuroimaging markers was carried out in accordance with the STRIVE recommendations. Cerebral Small Vessel Disease Score (CSVDS) was used to quantify the overall severity of MR imaging markers of CMA.Results.Among 70 examined (29 men and 41 women) aged 53.0 ± 14.2 years, average HD experience – 70.0 ± 39.5 months, the main clinical manifestations of CMA were cognitive impairment (82.9%, n = 58), emotional disorders (61.4%, n = 43), sleep disorders (38.6%, n = 27), pseudobulbar syndrome (17.1%, n = 12), walking disorders (8.6%, n = 6), acute lacunar syndromes (7.1%, n = 5) and pelvic dysfunction (4.3%, n = 3). CMA neuroimaging markers of varying severity were found in 100% of cases. Expansion of perivascular spaces (100%, n = 70) and white matter hyperintensities (81.4%, n = 57) prevailed in the structure of CMA imaging markers. Cortical atrophy (67%, n = 47), cerebral microbleeds (47%, n = 33), asymptomatic lacunae (35.7%, n = 25) and minor subcortical infarctions (2.9%, n = 3) were less common. Mild CMA (1–2 points on the CSVDS scale) was determined in 38 patients (54.3%), severe CMA (3–4 points on the CSVDS scale) – in 32 patients (45.7%). The presence of uncontrolled arterial hypertension (OR 1.85, p < 0.05), intradialysis hypertension (OR 2.8, p < 0.05), dialysis vegetative polyneuropathies (OR 2.75, p < 0.05), type 2 diabetes mellitus (OR 5.7, p < 0.05) and the experience of programmed HD (more than 50 months) (OR 3.1, p < 0.05) were prognostic signifi cance for the development of severe CMA in dialysis patients.Conclusion.All patients with end-stage CKD who have been on programmed HD for a long time are shown to undergo the brain MRI in order to timely diagnose CMA imaging markers and possible correction of therapy.

Establishing new approaches for the prevention and treatment of stroke relies on identifying modifiable risk factors that contribute to the development of this complex disease. Mendelian randomization (MR) studies, analogous to naturally occurring randomized trials, can assess causality of potentially modifiable biomarkers and offer new insights into biological pathways. Stroke is the second leading cause of death worldwide and the chief determinant of long-term disability. Stroke is a heterogeneous disease arising from several distinct underlying pathologies and is typically classified as ischemic or hemorrhagic, and further subclassified using imaging data. Ischemic stroke (IS), including its 3 main subtypes: small vessel disease, large vessel disease, and cardioembolic stroke, accounts for ≈80% of stroke and is the result of an interrupted blood supply, leading to localized areas of ischemia in the brain. Small vessel disease may be a consequence of nonatherosclerotic, as well as atherosclerotic, mechanisms that result in an occlusion of the small perforating arteries, whereas large vessel disease results from occlusions or emboli from plaque rupture in larger vessels, such as a carotid artery. Cardioembolic stroke arises typically from emboli from the heart. By contrast, hemorrhagic stroke is a consequence of intracerebral hemorrhage (bleeding into the brain) or subarachnoid hemorrhage (bleeding into the subarachnoid space). These diverse stroke subtypes have distinct underlying pathologies reflecting different risk factor distributions. MR studies, using genetic variants as instrumental variables, afford a powerful approach to assessing causality of risk factors and avoid biases inherent in observational studies, including confounding and reverse causation. This review considers the contribution of MR studies to stroke epidemiology and their relevance to understanding risk factors and new therapeutic targets for stroke. Meta-analyses of large prospective studies have enhanced our knowledge of classical and emerging risk factors for stroke.1–4 Classical risk factors for stroke include nonmodifiable characteristics, …

Both inflammation and cerebral white matter injury are closely associated with vascular cognitive impairment (VCI). The aim of this study was to analyze the correlation between peripheral serological markers, white matter injury, and cognitive function in patients with non-disabling ischemic cerebrovascular events (NICE); to identify potential biological markers for the diagnosis and prediction of VCI; and to provide a basis for the early diagnosis and intervention of VCI. We collected clinical data, along with demographic and medical history data, from 151 NICE patients. Fasting venous blood samples were collected. Based on the Montreal Cognitive Assessment (MoCA) after admission, we divided the patients into normal cognitive function (NCF) and VCI groups, and then classified them into mild white matter hyperintensity (mWMH) and severe white matter hyperintensity (sWMH) based on Fazekas scores. The differences in serological marker levels were compared between the cognitive function groups and the white matter hyperintensity groups. Binary logistic regression models and receiver operating characteristic curves were used to analyze the diagnostic predictive value of serological markers for VCI in patients with NICE and in the white matter hyperintensity subgroups. Among 151 patients with NICE, 95 were male and 56 were female. Lymphocyte count (OR = 0.405, p = 0.010, 95% CI [0.201, 0.806]), red blood cell count (OR = 0.433, p = 0.010, 95% CI [0.228, 0.821]), and hemoglobin level (OR = 0.979, p = 0.046, 95% CI [0.958, 0.999]) were protective factors for cognitive function in patients with NICE. The sWMH group had a higher age, granulocyte/lymphoid ratio (NLR), and neutrophil percentage but a lower MoCA score, hemoglobin level, and lymphocyte count than the mWMH group. In the mWMH group, lymphocyte count (AUC = 0.713, p = 0.003, 95% CI [0.593, 0.833]) had an acceptable predictive value for the diagnosis of VCI, whereas white blood cell count (AUC = 0.672, p = 0.011, 95% CI [0.545, 0.799]), red blood cell count (AUC = 0.665, p = 0.014, 95% CI [0.545, 0.784]), and hemoglobin level (AUC = 0.634, p = 0.047, 95% CI [0.502, 0.765]) had marginal predictive value for the diagnosis of VCI. In the sWMH group, no significant differences were found in serological markers between the NCF and VCI groups. Lymphocyte count, red blood cell count, and hemoglobin level were independent protective factors for cognitive function in patients with NICE; they can be used as potential biological markers to distinguish VCI in patients with NICE and are applicable to subgroups of patients with mWMH.

White matter hyperintensities (WMHs) are the commonest imaging marker of cerebral small vessel disease (SVD) and a major cause of cognitive decline and vascular dementia. WMHs typically accumulate over time, but recent studies show they can also regress, but potential clinical benefits have received little attention. We examined progressing, stable, and regressing WMH in people with stroke-related SVD and the effect on cognitive outcomes. We recruited patients with minor nondisabling ischemic stroke (modified Rankin score ≤2) from stroke services into our prospective longitudinal observational study. Participants underwent cognitive assessment and brain MRI within 3-month poststroke and 1 year later. We gathered information on vascular risk factors, stroke severity, global cognition (Montreal Cognitive Assessment [MoCA]), processing speed and executive functioning (Trail Making Test [TMT] A and B, and the B/A ratio with ratio ≥3 reflecting executive dysfunction), and the Letter Digit Substitution Test. We measured WMH volumes at baseline and 1 year and categorized net WMH volume change into quintiles: Q1 (most regression), Q3 (stable), and Q5 (most progression). We applied repeated-measures linear mixed models to analyze longitudinal WMH and cognitive changes, adjusting for age, sex, premorbid intelligence, stroke severity, disability, white matter structural integrity, and baseline WMH volume. One hundred ninety-eight of 229 participants had WMH volumes available at both time-points. At baseline, the mean age was 67.5 years (SD = 10.9), with 33% female. Mean net WMH volume change per quintile was Q1 -1.79 mL (SD = 1.54), Q2 -0.27 mL (0.20), Q3 0.35 mL (0.18), Q4 1.43 mL (0.48), and Q5 5.31 mL (3.07). MoCA deteriorated the most in participants with most WMH progression (Q5) (estimated β -0.428 [95% CI -0.750 to -0.106]), compared with stable WMH (Q3), with no clear deterioration in those with most WMH regression (Q1). TMT B/A ratio improved in participants with most WMH regression (Q1; -0.385 [-0.758 to -0.012]). WMH regression was associated with preserved global cognition and improved executive function, compared with stable WMH, while WMH progression was associated with global cognitive decline. Cognitive benefits of WMH regression suggest that WMH-affected tissue can recover, may explain variance in cognitive outcomes, offer an important intervention target, and should be assessed in other populations and longer follow-up times.

Cerebral small vessel disease is the most common cause of vascular cognitive impairment. It typically manifests with lacunar infarcts and ischaemic white matter lesions. However, little is known about how these lesions relate to the cognitive symptoms. Previous studies have found a poor correlation between the burden of ischaemic lesions and cognitive symptoms, thus leaving much of the variance in cognitive performance unexplained. The objective of the current study was to investigate the relationship between the location of subcortical ischaemic lesions and cognitive symptoms in small vessel disease. We applied a voxel-based lesion-symptom mapping approach to data from 215 patients with CADASIL, a genetically defined small vessel disease with mutations in the NOTCH3 gene. All patients were examined by magnetic resonance imaging and comprehensive neuropsychological testing. Lacunar lesions and white matter lesions were segmented on three-dimensional T(1) and fluid-attenuated inversion recovery sequences, respectively. One hundred and forty-five subjects had a total of 854 lacunar lesions (range 1-13 per individual). The normalized volume of white matter hyperintensities ranged from 0.0425% to 21.5% of the intracranial cavity. Significant clusters for cognitive performance were detected for both lacunar lesions and white matter hyperintensities. The most prominent results were obtained on a compound score for processing speed, the predominantly affected cognitive domain in this group of patients. Strategic locations included the anterior parts of the thalamus, the genu and anterior limb of the internal capsule, the anterior corona radiata and the genu of the corpus callosum. By combining the lesion-symptom mapping data with information from a probabilistic white matter atlas we found that the majority of the processing speed clusters projected on the anterior thalamic radiation and the forceps minor. In multivariate models that included demographic parameters, brain atrophy and the volume of ischaemic lesions, regional volumes of lacunar lesions and white matter hyperintensities in the anterior thalamic radiation predicted performance in processing speed tasks, whereas there was no independent contribution of the global volume of ischaemic lesions. These observations emphasize the importance of lesion location for both lacunar and ischaemic white matter lesions. Our findings further highlight the anterior thalamic radiation as a major anatomical structure impacting on processing speed. Together these findings provide strong support for a central role of frontal-subcortical circuits in cerebral small vessel disease and vascular cognitive impairment.

Objectives: The study aimed to analyze the brain white matter hyperintensities (WMHs) of patients with vascular dementia (VaD) and Alzheimer’s disease (AD) using magnetic resonance imaging to determine whether white matter lesions in the brain could be detected by a computer using image processing methods. Patients and methods: In this retrospective observational study, a unimodal, unsupervised, and automatic method was developed, and magnetic resonance imaging of 35 patients were examined. Of the 35 patients, 19 (14 males, 5 females; mean age: 73.2±6.7 years; range, 56 to 83 years) were picked from patients with AD or VaD who were admitted to a neurology clinic between January 2016 and December 2022 (Group 1). The remaining 16 patients (10 females, 5 males; mean age: 80.4±5.4 years; range, 69 to 92 years) were included from the ABVIB (Aging Brain: Vasculature, Ischemia, and Behavior) study from the ADNI (Alzheimer’s Disease Neuroimaging Initiative) database (Group 2). To calculate the volume of WMHs, a detailed analysis was conducted. Initially, skull stripping was performed, and then the brain was segmented. Afterward, two types of masks (Mask1 and Mask-2) were obtained by applying painting, decreasing, and blurring processes to the segmented white matter. These masks limited the region that was searched for WMHs. With this limitation, false positives that could arise from gray matter intensities were tried to be prevented. To evaluate the accuracy of WMH detection, a user interface was developed, and manual marking was conducted by an expert neurologist. After WMH detection, WMH volumes were calculated. Results: For Group 1, the similarity index was found to be 0.76 for Mask-1 and 0.80 for Mask-2, while for Group 2, the similarity index was 0.71 for Mask-1 and 0.87 for Mask-2. In patients with AD, the mean WMH lesion load (LL) was 15.16±16.59 mL. In patients with VaD, who were expected to suffer more from WMHs, the mean WMH LL was 29.22±11.40 mL. In Group 2, the mean WMH LL was 17.77±12.26 mL. Conclusion: This study may contribute to the literature since it is an automatic, unimodal, and unsupervised method that was applied to both a completely unique data set with many different scanning parameters and an open database data set.

Background. Stroke is a common cause of mortality and disability. There are many risk factors for stroke among which leukoaraiosis (LA) is mentioned. Historically, LA was a radiological term, however, today it is classified as Cerebral Small Vessel Disease (CSVD) which clinical presentation depends on the affected brain area. Higher prevalence of LA is found not only in stroke patients, but also in patients with hypertension and other cerebrovascular risk factors. Therefore, the aim of the study was to evaluate the relationship between the present LA, selected laboratory tests, the carotid ultrasound markers and cognitive tests results in patients with stroke. Material and methods. The study included 102 patients (W: 56, M: 46) at the age of 70.9 ± 11.5 hospitalized due to stroke in the Stroke Unit of the Department of Neurology. The clinical assessment included NIHSS score, MMSE testing, laboratory blood tests, carotid duplex USG and CT scan of the brain. Patients were dichotomized based on the presence of LA in the CT scan. Results. LA was present in 25 (24.5%) patients. It was more frequently found in older patients (> 72 years old; p < 0.001). In the LA group, higher levels of LDL cholesterol (p = 0.002), lower hemoglobin concentration (p = 0.03) and higher platelets count (p = 0.04) were observed. The carotid ultrasound showed higher intimamedia complexes in the LA group (p = 0.02). The functional test showed lower scores on the clock test in patients with LA (p = 0.04). The presence of LA was three times less likely to be present in patients administered with beta1-adrenolytics (p = 0.03). Conclusions. The occurrence of leukoaraiosis in patients with acute stroke is associated with clustering of other vascular risk factors cognitive impairment, and may be related to ongoing cardiovascular therapy.