A5059154016- Artificial Intelligence in Healthcare
- Machine Learning in Healthcare
- Artificial Intelligence in Healthcare and Education
- Topic Modeling
- Biomedical Text Mining and Ontologies
- Natural Language Processing Techniques
- COVID-19 diagnosis using AI
- Zoonotic diseases and public health
- Viral Infections and Outbreaks Research
- Genetics, Bioinformatics, and Biomedical Research
- SARS-CoV-2 detection and testing
- Electronic Health Records Systems
- Time Series Analysis and Forecasting
- COVID-19 Clinical Research Studies
- Adversarial Robustness in Machine Learning
- Autopsy Techniques and Outcomes
- Text Readability and Simplification
University of Oxford
2021-2024
Institute of Biomedical Science
2021-2022
Uncertainty in patients' COVID-19 status contributes to treatment delays, nosocomial transmission, and operational pressures hospitals. However, the typical turnaround time for laboratory PCR remains 12-24 h lateral flow devices (LFDs) have limited sensitivity. Previously, we shown that artificial intelligence-driven triage (CURIAL-1.0) can provide rapid screening using clinical data routinely available within 1 of arrival hospital. Here, aimed improve from emergency department availability...
Abstract Motivation Language models pre-trained on biomedical corpora, such as BioBERT, have recently shown promising results downstream tasks. Many existing models, the other hand, are resource-intensive and computationally heavy owing to factors embedding size, hidden dimension number of layers. The natural language processing community has developed numerous strategies compress these utilizing techniques pruning, quantization knowledge distillation, resulting in that considerably faster,...
Specialised pre-trained language models are becoming more frequent in Natural Processing (NLP) since they can potentially outperform trained on generic texts. BioBERT (Sanh et al., Distilbert, a distilled version of bert: smaller, faster, cheaper and lighter.
Abstract The adoption of large language models (LLMs) to assist clinicians has attracted remarkable attention. Existing works mainly adopt the closeended question-answering (QA) task with answer options for evaluation. However, many clinical decisions involve answering open-ended questions without pre-set options. To better understand LLMs in clinic, we construct a benchmark ClinicBench . We first collect eleven existing datasets covering diverse generation, understanding, and reasoning...
Irregular time series (ITS) are common in healthcare as patient data is recorded an electronic health record (EHR) system per clinical guidelines/requirements but not for research and depends on a patient's status. Due to irregularity, it challenging develop machine learning techniques uncover vast intelligence hidden EHR big data, without losing performance downstream outcome prediction tasks.
The adoption of large language models (LLMs) to assist clinicians has attracted remarkable attention. Existing works mainly adopt the close-ended question-answering task with answer options for evaluation. However, in real clinical settings, many decisions, such as treatment recommendations, involve answering open-ended questions without pre-set options. Meanwhile, existing studies use accuracy assess model performance. In this paper, we comprehensively benchmark diverse LLMs healthcare,...
Early detection of COVID-19 is an ongoing area research that can help with triage, monitoring and general health assessment potential patients may reduce operational strain on hospitals cope the coronavirus pandemic. Different machine learning techniques have been used in literature to detect cases using routine clinical data (blood tests, vital signs measurements). Data breaches information leakage when these models bring reputational damage cause legal issues for hospitals. In spite this,...
Abstract Irregular time-series (ITS) are prevalent in the electronic health records (EHR) as data is recorded EHR system per clinical guidelines/requirements but not for research and also depends on patient status. ITS present challenges training of machine learning algorithms, which mostly built assumption coherent fixed dimensional feature space. In this paper, we propose a computationally efficient variant transformer based idea cross-attention, called Perceiver, healthcare. We further...
<ns4:p>The COVID CIRCLE initiative Research Project Tracker by UKCDR and GloPID-R associated living mapping review (LMR) showed the importance of sharing analysing data on research at point funding to improve coordination during a pandemic. This approach can also help with preparedness for outbreaks hence our new programme Pandemic Preparedness: Analytical Capacity Funding Tracking Programme (Pandemic PACT) has been established. The LMR described in this protocol will provide an open,...
The COVID CIRCLE initiative Research Project Tracker by UKCDR and GloPID-R associated living mapping review (LMR) showed the importance of sharing analysing data on research at point funding to improve coordination during a pandemic. This approach can also help with preparedness for outbreaks hence our new programme Pandemic Preparedness: Analytical Capacity Funding Tracking Programme (Pandemic PACT) has been established. LMR described in this protocol builds previous COVID-19 database...
Specialised pre-trained language models are becoming more frequent in NLP since they can potentially outperform trained on generic texts. BioBERT and BioClinicalBERT two examples of such that have shown promise medical tasks. Many these overparametrised resource-intensive, but thanks to techniques like Knowledge Distillation (KD), it is possible create smaller versions perform almost as well their larger counterparts. In this work, we specifically focus development compact for processing...
Abstract Background Uncertainty in patients’ COVID-19 status contributes to treatment delays, nosocomial transmission, and operational pressures hospitals. However, typical turnaround times for batch-processed laboratory PCR tests remain 12-24h. Although rapid antigen lateral flow testing (LFD) has been widely adopted UK emergency care settings, sensitivity is limited. We recently demonstrated that AI-driven triage (CURIAL-1.0) allows high-throughput screening using clinical data routinely...