A5034420885- Topic Modeling
- Natural Language Processing Techniques
- Biomedical Text Mining and Ontologies
- Machine Learning in Healthcare
- AI in cancer detection
- Advanced Fluorescence Microscopy Techniques
- Speech Recognition and Synthesis
- Radiomics and Machine Learning in Medical Imaging
- Artificial Intelligence in Healthcare and Education
- Explainable Artificial Intelligence (XAI)
- Artificial Intelligence in Healthcare
- Advanced Neural Network Applications
- Heart Rate Variability and Autonomic Control
- melanin and skin pigmentation
- Brain Metastases and Treatment
- Human-Automation Interaction and Safety
- Optical Coherence Tomography Applications
- Brain Tumor Detection and Classification
- Medical Image Segmentation Techniques
- Non-Invasive Vital Sign Monitoring
- Neuroscience and Neural Engineering
- ECG Monitoring and Analysis
- Sleep and Work-Related Fatigue
- Healthcare professionals’ stress and burnout
- Cutaneous Melanoma Detection and Management
Google (United States)
2024
Massachusetts Institute of Technology
2018-2023
IIT@MIT
2021
National Taiwan University
2015-2021
Harvard University
2016-2020
Massachusetts General Hospital
2020
University of California, San Francisco
2020
IBM (United States)
2020
Vassar College
2019
Chang Gung University
2011-2016
Contextual word embedding models such as ELMo and BERT have dramatically improved performance for many natural language processing (NLP) tasks in recent months. However, these been minimally explored on specialty corpora, clinical text; moreover, the domain, no publicly-available pre-trained yet exist. In this work, we address need by exploring releasing text: one generic text another discharge summaries specifically. We demonstrate that using a domain-specific model yields improvements 3/5...
Contextual word embedding models such as ELMo (Peters et al., 2018) and BERT (Devlin have dramatically improved performance for many natural language processing (NLP) tasks in recent months. However, these been minimally explored on specialty corpora, clinical text; moreover, the domain, no publicly-available pre-trained yet exist. In this work, we address need by exploring releasing text: one generic text another discharge summaries specifically. We demonstrate that using a domain-specific...
Survival outcome prediction is a challenging weakly-supervised and ordinal regression task in computational pathology that involves modeling complex interactions within the tumor microenvironment gigapixel whole slide images (WSIs). Despite recent progress formulating WSIs as bags for multiple instance learning (MIL), representation of entire remains an open problem, especially overcoming: 1) complexity feature aggregation large bags, 2) data heterogeneity gap incorporating biological priors...
Open domain question answering (OpenQA) tasks have been recently attracting more and attention from the natural language processing (NLP) community. In this work, we present first free-form multiple-choice OpenQA dataset for solving medical problems, MedQA, collected professional board exams. It covers three languages: English, simplified Chinese, traditional contains 12,723, 34,251, 14,123 questions languages, respectively. We implement both rule-based popular neural methods by sequentially...
Abstract Two-dimensional materials such as graphene have shown great promise biosensors, but suffer from large device-to-device variation due to non-uniform material synthesis and device fabrication technologies. Here, we develop a robust bioelectronic sensing platform composed of more than 200 integrated units, custom-built high-speed readout electronics, machine learning inference that overcomes these challenges achieve rapid, portable, reliable measurements. The demonstrates...
Excellence in a wide variety of medical applications poses considerable challenges for AI, requiring advanced reasoning, access to up-to-date knowledge and understanding complex multimodal data. Gemini models, with strong general capabilities long-context offer exciting possibilities medicine. Building on these core strengths Gemini, we introduce Med-Gemini, family highly capable models that are specialized medicine the ability seamlessly use web search, can be efficiently tailored novel...
Over 85 million computed tomography (CT) scans are performed annually in the US, of which approximately one quarter focus on abdomen. Given current shortage both general and specialized radiologists, there is a large impetus to use artificial intelligence alleviate burden interpreting these complex imaging studies while simultaneously using images extract novel physiological insights. Prior state-of-the-art approaches for automated medical image interpretation leverage vision language models...
The medical subdomain of a clinical note, such as cardiology or neurology, is useful content-derived metadata for developing machine learning downstream applications. To classify the note accurately, we have constructed learning-based natural language processing (NLP) pipeline and developed classifiers based on content note. We using NLP system, Text Analysis Knowledge Extraction System (cTAKES), Unified Medical Language (UMLS) Metathesaurus, Semantic Network, algorithms to extract features...
The automatic generation of radiology reports given medical radiographs has significant potential to operationally and improve clinical patient care. A number prior works have focused on this problem, employing advanced methods from computer vision natural language produce readable reports. However, these often fail account for the particular nuances domain, and, in particular, critical importance accuracy resulting generated In work, we present a domain-aware chest X-ray report system which...
Deep neural networks have been investigated in learning latent representations of medical images, yet most the studies limit their approach a single supervised convolutional network (CNN), which usually rely heavily on large scale annotated dataset for training. To learn image with less supervision involved, we propose deep Siamese CNN (SCNN) architecture that can be trained only binary pair information. We evaluated learned task content-based retrieval using publicly available multiclass...
Open domain question answering (OpenQA) tasks have been recently attracting more and attention from the natural language processing (NLP) community. In this work, we present first free-form multiple-choice OpenQA dataset for solving medical problems, MedQA, collected professional board exams. It covers three languages: English, simplified Chinese, traditional contains 12,723, 34,251, 14,123 questions languages, respectively. We implement both rule-based popular neural methods by sequentially...
Many clinical tasks require an understanding of specialized data, such as medical images and genomics, which is not typically found in general-purpose large multimodal models. Building upon Gemini's models, we develop several models within the new Med-Gemini family that inherit core capabilities Gemini are optimized for use via fine-tuning with 2D 3D radiology, histopathology, ophthalmology, dermatology genomic data. Med-Gemini-2D sets a standard AI-based chest X-ray (CXR) report generation...
Recent research has shown that word embedding spaces learned from text corpora of different languages can be aligned without any parallel data supervision. Inspired by the success in unsupervised cross-lingual embeddings, this paper we target learning a cross-modal alignment between speech and their respective modalities an fashion. The proposed framework learns individual spaces, attempts to align two via adversarial training, followed refinement procedure. We show how our could used...
AI models have been proposed for hypothesis generation, but testing their ability to drive high-impact research is challenging, since an AI-generated can take decades validate. Here, we challenge the of a recently developed LLM-based platform, co-scientist, generate high-level hypotheses by posing question that took years resolve experimentally remained unpublished: How could capsid-forming phage-inducible chromosomal islands (cf-PICIs) spread across bacterial species? Remarkably,...
Glycemic control is essential for critical care. However, it a challenging task because there has been no study on personalized optimal strategies glycemic control. This work aims to learn trajectories severely ill septic patients by learning data-driven policies identify targeted blood glucose levels as reference clinicians. We encoded patient states using sparse autoencoder and adopted reinforcement paradigm policy iteration the from data. also estimated expected return following learned...
We present a framework for building speech-to-text translation (ST) systems using only monolingual speech and text corpora, in other words, utterances from source language independent target language. As opposed to traditional cascaded end-to-end architectures, our system does not require any labeled data (i.e., transcribed audio or parallel corpora) during training, making it especially applicable pairs with very few even zero bilingual resources. The initializes the ST cross-modal...
In this work, we present an approach, which call Embeddings for Language/Image-aligned X-Rays, or ELIXR, that leverages a language-aligned image encoder combined grafted onto fixed LLM, PaLM 2, to perform broad range of chest X-ray tasks. We train lightweight adapter architecture using images paired with corresponding free-text radiology reports from the MIMIC-CXR dataset. ELIXR achieved state-of-the-art performance on zero-shot (CXR) classification (mean AUC 0.850 across 13 findings),...
Health acoustic sounds such as coughs and breaths are known to contain useful health signals with significant potential for monitoring disease, yet underexplored in the medical machine learning community. The existing deep systems acoustics often narrowly trained evaluated on a single task, which is limited by data may hinder generalization other tasks. To mitigate these gaps, we develop HeAR, scalable self-supervised learning-based system using masked autoencoders large dataset of 313...
Internship, the transition period from medical student to junior doctor, is highly stressful for interns in West; however, little known about experience of coping with stress Taiwan. This study aimed develop a model among Taiwanese and examine relationship between learning outcomes. For this qualitative study, we used grounded theory methodology theoretical sampling. We collected data through in-depth interviews participant observations. employed constant comparative method analyse until...