A5101413400- Neural dynamics and brain function
- Face Recognition and Perception
- Visual perception and processing mechanisms
- Cardiac Imaging and Diagnostics
- Cardiovascular Function and Risk Factors
- Domain Adaptation and Few-Shot Learning
- Ultrasound in Clinical Applications
- Neural Networks and Applications
- Artificial Intelligence in Healthcare and Education
- Visual Attention and Saliency Detection
- Energy Efficient Wireless Sensor Networks
- Cardiac Valve Diseases and Treatments
- Non-Invasive Vital Sign Monitoring
- Wireless Networks and Protocols
- Advanced X-ray and CT Imaging
- Radiation Dose and Imaging
- Blood properties and coagulation
- Colorectal Cancer Screening and Detection
- Radiology practices and education
- Surgical Simulation and Training
- Medical Image Segmentation Techniques
- Indoor and Outdoor Localization Technologies
- Cardiac, Anesthesia and Surgical Outcomes
- Advanced Memory and Neural Computing
- Physics and Engineering Research Articles
Medtronic (United States)
2023-2025
Harvard–MIT Division of Health Sciences and Technology
2014-2021
McGovern Institute for Brain Research
2013-2021
Massachusetts Institute of Technology
1997-2021
Northwestern University
2020-2021
Bay Institute
2018-2020
Stanford University
2019
MedStar Health
2019
Institute of Cognitive and Brain Sciences
2015-2016
University of Cambridge
2009
Significance Humans and monkeys easily recognize objects in scenes. This ability is known to be supported by a network of hierarchically interconnected brain areas. However, understanding neurons higher levels this hierarchy has long remained major challenge visual systems neuroscience. We use computational techniques identify neural model that matches human performance on challenging object categorization tasks. Although not explicitly constrained match data, turns out highly predictive...
The primate visual system achieves remarkable object recognition performance even in brief presentations, and under changes to exemplar, geometric transformations, background variation (a.k.a. core recognition). This is mediated by the representation formed inferior temporal (IT) cortex. In parallel, recent advances machine learning have led ever higher performing models of using artificial deep neural networks (DNNs). It remains unclear, however, whether representational DNNs rivals that...
The internal representations of early deep artificial neural networks (ANNs) were found to be remarkably similar the measured experimentally in primate brain. Here we ask, as ANNs have continued evolve, are they becoming more or less brain-like? that most functionally brain will contain mechanisms like those used by We therefore developed Brain-Score – a composite multiple and behavioral benchmarks score any ANN on how it is brain’s for core object recognition deployed evaluate wide range...
Artificial intelligence (AI) has been applied to analysis of medical imaging in recent years, but AI guide the acquisition ultrasonography images is a novel area investigation. A deep-learning (DL) algorithm, trained on more than 5 million examples outcome ultrasonographic probe movement image quality, can provide real-time prescriptive guidance for novice operators obtain limited diagnostic transthoracic echocardiographic images.
To go beyond qualitative models of the biological substrate object recognition, we ask: can a single ventral stream neuronal linking hypothesis quantitatively account for core recognition performance over broad range tasks? We measured human in 64 tests using thousands challenging images that explore shape similarity and identity preserving variation. then used multielectrode arrays to measure population responses those same visual areas V4 inferior temporal (IT) cortex monkeys simulated V1...
Background: Echocardiographic quantification of left ventricular (LV) ejection fraction (EF) relies on either manual or automated identification endocardial boundaries followed by model-based calculation end-systolic and end-diastolic LV volumes. Recent developments in artificial intelligence resulted computer algorithms that allow near detection measurement volumes function. However, boundary is still prone to errors limiting accuracy certain patients. We hypothesized a fully machine...
Deep convolutional artificial neural networks (ANNs) are the leading class of candidate models mechanisms visual processing in primate ventral stream. While initially inspired by brain anatomy, over past years, these ANNs have evolved from a simple eight-layer architecture AlexNet to extremely deep and branching architectures, demonstrating increasingly better object categorization performance, yet bringing into question how brain-like they still are. In particular, typical machine learning...
We have recently tested an automated machine-learning algorithm that quantifies left ventricular (LV) ejection fraction (EF) from guidelines-recommended apical views. However, in the point-of-care (POC) setting, 2-chamber views are often difficult to obtain, limiting usefulness of this approach. Since most POC physicians rely on visual assessment 4-chamber and parasternal long-axis views, our was adapted use either one these 3 or any combination. This study aimed (1) test accuracy estimates;...
Unsupervised representation learning has significantly advanced various machine tasks. In the computer vision domain, state-of-the-art approaches utilize transformations like random crop and color jitter to achieve invariant representations, embedding semantically same inputs despite transformations. However, this can degrade performance in tasks requiring precise features, such as localization or flower classification. To address this, recent research incorporates equivariant learning,...
Temporal continuity of object identity is a feature natural visual input and potentially exploited – in an unsupervised manner by the ventral stream to build neural representation inferior temporal (IT) cortex. Here, we investigated whether plasticity individual IT neurons underlies human core recognition behavioral changes induced with experience. We built single-neuron model combined previously established population-to-recognition-behavior-linking predict learning effects. found that our...
Abstract Human visual object recognition is subserved by a multitude of cortical areas. To make sense this system, one line research focused on response properties primary cortex neurons and developed theoretical models set canonical computations such as convolution, thresholding, exponentiating normalization that could be hierarchically repeated to give rise more complex representations. Another or high-level linked these semantic categories useful for recognition. Here, we hypothesized the...
A key requirement for the development of effective learning representations is their evaluation and comparison to we know be effective. In natural sensory domains, community has viewed brain as a source inspiration an implicit benchmark success. However, it not been possible directly test representational algorithms against contained in neural systems. Here, propose new visual on which have tested representation multiple cortical areas macaque (utilizing data from [Majaj et al., 2012]), any...
Visual and auditory cortex both support impressively robust invariant recognition abilities, but operate on distinct classes of signals. To what extent are similar computations used across modalities? We examined this question by comparing state-of-the-art computational models to neural data from visual cortex. Using recent “deep learning” techniques, we built two hierarchical convolutional networks: an network optimized recognize words spectrograms, a categorize objects images. Each...
Abstract Funding Acknowledgements Bay Labs, Inc; San Francisco, CA Background/Introduction: When used by experienced examiners, the utility of point-of-care (POC) ultrasound for assessing cardiac anatomy and function has been well established. However, in some clinical circumstances (Primary Care offices, Intensive Unit, Emergency Rooms, or remote settings) which a rapid assessment dynamics can facilitate patient care, an examiner at POC scanning may not be immediately available. Purpose To...