A5101500968- Face and Expression Recognition
- Face recognition and analysis
- Human Pose and Action Recognition
- Video Surveillance and Tracking Methods
- Anomaly Detection Techniques and Applications
- Machine Learning in Healthcare
- Video Analysis and Summarization
- Emotion and Mood Recognition
- Neural Networks and Applications
- COVID-19 Clinical Research Studies
- Artificial Intelligence in Healthcare and Education
- Retinal Diseases and Treatments
- Domain Adaptation and Few-Shot Learning
- Stroke Rehabilitation and Recovery
- Advanced Neural Network Applications
- Visual Attention and Saliency Detection
- Image Retrieval and Classification Techniques
- Time Series Analysis and Forecasting
- Diabetes Management and Research
- Hand Gesture Recognition Systems
- Multimedia Communication and Technology
- Multimodal Machine Learning Applications
- Dermatological and COVID-19 studies
- 3D Shape Modeling and Analysis
- Advanced Image and Video Retrieval Techniques
Stanford University
2020-2025
Kennesaw State University
2017-2024
Singapore General Hospital
2024
Dunedin Public Hospital
2024
VinUniversity
2023
Stony Brook University
2016-2023
State University of New York
2022-2023
Carnegie Mellon University
2008-2020
Massachusetts General Hospital
2020
Harvard University
2020
Current methods of assessing psychopathology depend almost entirely on verbal report (clinical interview or questionnaire) patients, their family, caregivers. They lack systematic and efficient ways incorporating behavioral observations that are strong indicators psychological disorder, much which may occur outside the awareness either individual. We compared clinical diagnosis major depression with automatically measured facial actions vocal prosody in patients undergoing treatment for...
Visual categorization problems, such as object classification or action recognition, are increasingly often approached using a detection strategy: classifier function is first applied to candidate subwindows of the image video, and then maximum score used for class decision. Traditionally, subwindow classifiers trained on large collection examples manually annotated with masks bounding boxes. The reliance time-consuming human labeling effectively limits application these methods problems...
0. Abstract Background The integration of large language models (LLMs) in healthcare offers immense opportunity to streamline tasks, but also carries risks such as response accuracy and bias perpetration. To address this, we conducted a red-teaming exercise assess LLMs developed dataset clinically relevant scenarios for future teams use. Methods We convened 80 multi-disciplinary experts evaluate the performance popular across multiple medical scenarios. Teams composed clinicians, engineering...
Automatic facial action unit (AU) detection from video is a long-standing problem in computer vision. Two main approaches have been pursued: (1) static modeling - typically posed as discriminative classification which each frame evaluated independently; (2) temporal frames are segmented into sequences and modeled with variant of dynamic Bayesian networks. We propose segment-based approach, kSeg-SVM, that incorporates benefits both avoids their limitations. kSeg-SVM extension the spatial...
We present Hand-CNN, a novel convolutional network architecture for detecting hand masks and predicting orientations in unconstrained images. Hand-CNN extends MaskRCNN with attention mechanism to incorporate contextual cues the detection process. This can be implemented as an efficient module that captures non-local dependencies between features. inserted at different stages of object network, entire detector trained end-to-end. also introduce large-scale annotated datasets containing hands...
Abstract Objective The study sought to determine whether machine learning can predict initial inpatient total daily dose (TDD) of insulin from electronic health records more accurately than existing guideline-based dosing recommendations. Materials and Methods Using a tertiary academic center between 2008 2020 16,848 inpatients receiving subcutaneous who achieved target blood glucose control 100-180 mg/dL on calendar day, we trained an ensemble algorithm consisting regularized regression,...
Red teaming, the practice of adversarially exposing unexpected or undesired model behaviors, is critical towards improving equity and accuracy large language models, but non-model creator-affiliated red teaming scant in healthcare. We convened teams clinicians, medical engineering students, technical professionals (80 participants total) to stress-test models with real-world clinical cases categorize inappropriate responses along axes safety, privacy, hallucinations/accuracy, bias. Six...
Parameterized Appearance Models (PAMs) (e.g. eigen-tracking, active appearance models, morphable models) use Principal Component Analysis (PCA) to model the shape and of objects in images. Given a new image with an unknown appearance/shape configuration, PAMs can detect track object by optimizing model's parameters that best match image. While have numerous advantages for registration relative alternative approaches, they suffer from two major limitations: First, PCA cannot non-linear...
Abstract Many categories of objects, such as human faces, can be naturally viewed a composition several different layers. For example, bearded face with glasses decomposed into three layers: layer for glasses, the beard and other permanent facial features. While modeling linear subspace model could very difficult, separation allows easy modification some certain structures while leaving others unchanged. In this paper, we present method automatic extraction its applications to synthesis...
Automatic facial feature localization has been a long-standing challenge in the field of computer vision for several decades. This can be explained by large variation face an image have due to factors such as position, expression, pose, illumination, and background clutter. Support Vector Machines (SVMs) popular statistical tool detection. Traditional SVM approaches detection typically extract features from images (e.g. multiband filter, SIFT features) learn parameters. Independently...
Pretreatment positron emission tomography (PET) with 2-deoxy-2-[
Parameterized Appearance Models (PAMs) (e.g. Eigentracking, Active Models, Morphable Models) are commonly used to model the appearance and shape variation of objects in images. While PAMs have numerous advantages relative alternate approaches, they at least two drawbacks. First, especially prone local minima fitting process. Second, often few if any cost function correspond acceptable solutions. To solve these problems, this paper proposes a method learn by explicitly optimizing that occur...
Abstract Objective To develop prediction models for intensive care unit (ICU) vs non-ICU level-of-care need within 24 hours of inpatient admission emergency department (ED) patients using electronic health record data. Materials and Methods Using records 41 654 ED visits to a tertiary academic center from 2015 2019, we tested 4 algorithms—feed-forward neural networks, regularized regression, random forests, gradient-boosted trees—to predict ICU at the 24th hour following admission....
BACKGROUND Massive transfusion protocols to treat postinjury hemorrhage are based on predefined blood product ratios followed by goal-directed patient's clinical evolution. However, it remains unclear how these impact patient outcomes over time from injury. METHODS The Pragmatic, Randomized Optimal Platelet and Plasma Ratios (PROPPR) is a phase 3, randomized controlled trial, across 12 Level I trauma centers in North America. From 2012 2013, 680 severely injured patients required massive...
The National Synchrotron Light Source II (NSLS-II) at Brookhaven Laboratory (BNL) is now providing some of the world's brightest x-ray beams. A suite imaging and diffraction methods, exploiting megapixel detectors with kilohertz frame-rates NSLS-II beamlines, generate a variety image streams in unprecedented velocities volumes. complete understanding complex material system often requires cluster characterization tools that can reveal its elemental, structural, chemical physical properties...
In this paper, we tackle the problem of Crowd Counting, and present a crowd density estimation based approach for obtaining count. Most existing counting approaches rely on local features estimating map. work, investigate usefulness combining with non-local counting. We use convolution layers extracting features, type self-attention mechanism features. combine it conduct experiments three publicly available Counting datasets, achieve significant improvement over previous approaches.