Emily and Sam will present the work Subgraph Neural Networks at NeurIPS 2020|
Geeticka will present the work Joint Modeling of Chest Radiographs and Radiology Reports for Pulmonary Edema Assessment at MICCAI 2020|
Matthew presented the work CheXpert++: Approximating the CheXpert Labeler for Speed, Differentiability, and Probabilistic Output at MLHC 2020|
Di has completed the PhD thesis defense. Congratulations!|
Harry and Pete have the work Three-Dimensional Neural Network to Automatically Assess Liver Tumor Burden Change on Consecutive Liver MRIs published at Journal of the American College of Radiology|
Sam has completed the PhD thesis defense. Congratulations!|
Wei-Hung and Pete have the work Entity-Enriched Neural Models for Clinical Question Answering published at ACL BioNLP 2020|
The Clinical Decision Making Group at the MIT Laboratory for Computer Science is a research group dedicated to exploring and furthering the application of technology and artificial intelligence to clinical situations. Because of the vital and crucial nature of medical practice, and the need for accurate and timely information to support clinical decisions, the group is also focused on the gathering, availability, security and use of medical information throughout the human "life cycle" and beyond.
We collaborate closely with doctors and biomedical scientists. We derive knowledge from the growing set of patients' health records that note the natural histories of diseases and the outcomes of clinical care interventions. Such models can predict outcomes in new cases depending on what actions are chosen. Optimizing such choices enables recommendations for care that has the highest expected benefit.
As part of our approach, we use natural language processing methods to extract meaningful data from the clinical narratives that contain most observations made by doctors, nurses and other specialists. We apply a variety of machine learning techniques, including deep learning, matrix and tensor factorization, Gaussian processes, support vector machines, conditional random fields, logistic regression, Bayesian models, random forests, and reinforcement learning, and adapt them to the peculiar characteristics of clinical data. These include irregular sampling of data and observations, data that are not missing at random, narratives that are full of duplication, ambiguities and telegraphic abbreviations and elisions, data recorded by multiple instruments and people that are not well calibrated, etc.
We also work on related issues such as using personal health information systems to engage patients and families in care, augmenting classical clinical data sets with data from wearable instrumentation, social media reports, environmental exposures and the vastly expanding sets of genetic and genomic data. We have also contributed to advances in protecting patient privacy while allowing use of records for research, development of more flexible and supportive informational tools and interfaces, use of speech to interact with systems, etc.