Type of Position: Research Fellowship (Bolsa de Investigação)
Type of Contract: Research grant
Duration: 5 Months
Closed at: 2020-Dec-15
ICare4U is a project that proposes the development of a decision support system based on intelligent modelling and patient sub-group analysis, to provide personalized therapy for critically ill patients. It is hypothesized that the 'Collective Experience' from large clinical databases, where clinical decisions are linked with patient outcomes, can be used to identify specific patient sub-groups and build personalized therapy models towards a new era of personalized medicine, allowing the improvement of patient outcomes in the ICU. The validity of the proposed systems will be tested using two case studies where generalized severity scoring systems have consistently performed poorly: patients admitted to the ICU who then develop acute kidney injury, where two-thirds of these patients require renal support therapy; and severe sepsis a typical heterogeneous disease among critically ill patients, in which the immune response is highly dynamic and variable, and likely to need complex therapies to improve patient outcomes. One of the tasks of the project involves using existing text-based information, such as medical reports and physicians and nurses’ notes, to improve the decision-making process. The work to be developed by the researcher will be in the context of this task, and consists of implementing and integrating a technique called “Fuzzy Fingerprints” in the decision support system, and to enhance and integrate Fuzzy Fingerprints with Deep Learning. Digital fingerprints aim to extract information about a user that he/she has not provided knowingly, and represent such information in a compact and efficient form. Fuzzy fingerprints based on the order of top-k events have been successfully applied in many areas, including text mining, and will be used in this project to help identifying critically ill patients (by creating patient sub-group fingerprints). The technique presents a simple but extremely effective and robust way to evaluate and compare the relevant features of event patterns, which in the present case would be the text contained in the physicians/nurses notes. Important phrases and/or keywords will be used as events and event patterns. A text based fingerprint will be created for each admitted patient. A sliding window top-k algorithm will be adapted to create an evolving patient fingerprint algorithm to cope with patients’ evolving condition. Patients will be attributed to patient sub-groups based on fuzzy fingerprint similarity. The MIMIC II database will be used for this project. All text data will have to be “cleaned up” on a pre-processing phase since textual medical databases have several particularities that prevent its immediate and effective use.
João Paulo Baptista de Carvalho