A5063364372- Machine Learning in Healthcare
- Time Series Analysis and Forecasting
- Cardiac Valve Diseases and Treatments
- Advanced Neural Network Applications
- Machine Learning and Data Classification
- Healthcare Technology and Patient Monitoring
- Non-Invasive Vital Sign Monitoring
- Artificial Intelligence in Healthcare
- Infective Endocarditis Diagnosis and Management
- Video Surveillance and Tracking Methods
- Dementia and Cognitive Impairment Research
- Digital Mental Health Interventions
- Brain Tumor Detection and Classification
- Medical Image Segmentation Techniques
- Anomaly Detection Techniques and Applications
- Visual Attention and Saliency Detection
- Cancer-related molecular mechanisms research
- demographic modeling and climate adaptation
- Emotion and Mood Recognition
- Music and Audio Processing
- Genetic Associations and Epidemiology
- Advanced Image and Video Retrieval Techniques
- Neural Networks and Applications
- Reinforcement Learning in Robotics
- Algorithms and Data Compression
University of Massachusetts Amherst
2019-2025
Amherst College
2022-2025
University of Massachusetts Boston
2023
Stanford University
2015-2020
Mobilize
2018
Laboratoire d'Informatique de Paris-Nord
2017
Carnegie Mellon University
2012-2016
University of Pittsburgh
2013
Polytechnic University of Timişoara
2009
We present Deep Neural Decision Forests - a novel approach that unifies classification trees with the representation learning functionality known from deep convolutional networks, by training them in an end-to-end manner. To combine these two worlds, we introduce stochastic and differentiable decision tree model, which steers usually conducted initial layers of (deep) network. Our model differs conventional networks because forest provides final predictions it forests since propose...
Abstract The primary practice of healthcare artificial intelligence (AI) starts with model development, often using state-of-the-art AI, retrospectively evaluated metrics lifted from the AI literature like AUROC and DICE score. However, good performance on these may not translate to improved clinical outcomes. Instead, we argue for a better development pipeline constructed by working backward end goal positively impacting clinically relevant outcomes leading considerations causality in...
Thermal images are mainly used to detect the presence of people at night or in bad lighting conditions, but perform poorly daytime. To solve this problem, most state-of-the-art techniques employ a fusion network that uses features from paired thermal and color images. Instead, we propose augment with their saliency maps, serve as an attention mechanism for pedestrian detector especially during We investigate how such approach results improved performance detection using only images,...
Biomedical repositories such as the UK Biobank provide increasing access to prospectively collected cardiac imaging, however these data are unlabeled, which creates barriers their use in supervised machine learning. We develop a weakly deep learning model for classification of aortic valve malformations using up 4,000 unlabeled MRI sequences. Instead requiring highly curated training data, weak supervision relies on noisy heuristics defined by domain experts programmatically generate...
Objective: The use of machine-learning algorithms to classify alerts as real or artifacts in online noninvasive vital sign data streams reduce alarm fatigue and missed true instability. Design: Observational cohort study. Setting: Twenty-four–bed trauma step-down unit. Patients: Two thousand one hundred fifty-three patients. Intervention: Noninvasive monitoring (heart rate, respiratory peripheral oximetry) recorded on all admissions at 1/20 Hz, blood pressure less frequently, partitioned...
<h3>Importance</h3> The use of machine learning applications related to health is rapidly increasing and may have the potential profoundly affect field care. <h3>Objective</h3> To analyze submissions a popular for venue assess current state research, including areas methodologic clinical focus, limitations, underexplored areas. <h3>Design, Setting, Participants</h3> In this data-driven qualitative analysis, 166 accepted manuscript Third Annual Machine Learning Health workshop at 32nd...
In intensive care units (ICUs), patient health is monitored through (1) continuous vital signals from various medical devices, and (2) clinical notes consisting of opinions summaries doctors which are recorded in electronic records (EHR).It difficult to jointly model these two sources information because notes, unlike signals, collected at irregular intervals their contents relatively unstructured.In this paper, we present a that combines both about ICU patients make accurate in-hospital...
Background: The aortic valve is an important determinant of cardiovascular physiology and anatomic location common human diseases. Methods: From a sample 34 287 white British ancestry participants, we estimated functional area by planimetry from prospectively obtained cardiac magnetic resonance imaging sequences the valve. Aortic measurements were submitted to genome-wide association testing, followed polygenic risk scoring phenome-wide screening, identify genetic comorbidities. Results: A...
The fibrous annulus of the mitral valve plays an important role in valvular function and cardiac physiology, while normal variation size cardiovascular anatomy may share a genetic link with common rare disease. We derived automated estimates annular diameter 4-chamber view from 32,220 MRI images UK Biobank at ventricular systole diastole as basis for GWAS. Mitral dimensions corresponded to previously described anatomical norms, GWAS inclusive 4 population strata identified 10 loci, including...
Predicting cancer drug response using both genomics and features has shown some success compared to alone. However, there been limited research done on how best combine or fuse the two types of features. Using a visible neural network with deep learning branches for genes as base architecture, we experimented different fusion functions points. Our experiments show that injecting multiplicative relationships between gene latent into original concatenation-based architecture DrugCell...
Abstract Biomedical repositories such as the UK Biobank provide increasing access to prospectively collected cardiac imaging, however these data are unlabeled which creates barriers their use in supervised machine learning. We develop a weakly deep learning model for classification of aortic valve malformations using up 4,000 MRI sequences. Instead requiring highly curated training data, weak supervision relies on noisy heuristics defined by domain experts programmatically generate...
In recent years, many papers have reported state-of-the-art performance on Alzheimer's Disease classification with MRI scans from the Neuroimaging Initiative (ADNI) dataset using convolutional neural networks. However, we discover that when split data into training and testing sets at subject level, are not able to obtain similar performance, bringing validity of previous studies question. Furthermore, point out works use different subsets ADNI data, making comparison across tricky. this...
The COVID-19 pandemic has significantly impacted academic life in the United States and beyond. To gain a better understanding of its impact on community, we conducted large-scale survey at University Massachusetts Amherst. We collected multifaceted data from students, staff, faculty several aspects their lives, such as mental physical health, productivity, finances. All our respondents expressed issues concerns, increased stress depression levels. Financial difficulties seem to have most...
A plethora of deep learning models have been developed for the task Alzheimer's disease classification from brain MRI scans. Many these report high performance, achieving three-class accuracy up to 95%. However, it is common studies draw performance comparisons between that are trained on different subsets a dataset or use varying imaging preprocessing techniques, making difficult objectively assess model performance. Furthermore, many works do not provide details such as hyperparameters,...
In healthcare applications, temporal variables that encode movement, health status and longitudinal patient evolution are often accompanied by rich structured information such as demographics, diagnostics medical exam data. However, current methods do not jointly optimize over covariates time series in the feature extraction process. We present ShortFuse, a method boosts accuracy of deep learning models for explicitly modeling interactions dependencies with covariates. ShortFuse introduces...
Abstract College students experience ever-increasing levels of stress, leading to a wide range health problems. In this context, monitoring and predicting students’ stress is crucial and, fortunately, made possible by the growing support for data collection via mobile devices. However, from phone remains challenging task, off-the-shelf deep learning models are inapplicable or inefficient due irregularity, inter-subject variability, “cold start problem”. To overcome these challenges, we...
This volume represents the accepted submissions from Machine Learning for Health (ML4H) workshop at conference on Neural Information Processing Systems (NeurIPS) 2018, held December 8, 2018 in Montreal, Canada.
Integration of multimodal information from various sources has been shown to boost the performance machine learning models and thus received increased attention in recent years. Often such use deep modality-specific networks obtain unimodal features which are combined "late-fusion" representations. However, these designs run risk loss respective pipelines. On other hand, "early-fusion" methodologies, combine early, suffer problems associated with feature heterogeneity high sample complexity....
With the growing popularity of wearable devices, ability to utilize physiological data collected from these devices predict wearer's mental state such as mood and stress suggests great clinical applications, yet a task is extremely challenging. In this paper, we present general platform for personalized predictive modeling behavioural states like students' level stress. Through use Auto-encoders Multitask learning extend prediction both sequences passive sensor high-level covariates. Our...