Automated Structuring for Neuro-oncology
An appropriate review of a patient’s medical record often requires that a physician review multiple clinical documents while mentally noting issues related to the case. Physicians must review what the findings were, the chronology of events, spatial/temporal patterns of disease progression, the effects of interventions and the possible causal lines of explanation of observed findings. Among all this data, the physician must filter out information irrelevant to the current clinical context of care. And unfortunately, imaging data and image-derived conclusions are typically poorly integrated into patient care and management. The growing complexity of the medical record makes it more difficult for physicians to perform a comprehensive review of the patient chart, which may compromise care in several ways, including: lack of caregiver coordination, poor integration of exam results, initiating treatment before diagnosis is established, administration of inappropriate therapies, and/or performing redundant studies.
This work addressed the development of a system for facilitating the review of clinical patient data intended to promote an orderly process of medical problem understanding and care. Phenomenological models of disease were used with natural language processing (NLP) to abstract information from medical reports. An open source toolkit to allow researchers to train the NLP system for the specific domain of neurooncology was developed. In addition, a front-end application for visualizing text and imaging data to improve the understanding of underlying data patterns in the medical record was created.