Ontology based Image Analysis for Clinical Care Pathways

In oncology treatment, clinical care pathways play a major role in the prognosis of the disease. Based on the significant research collaboration between academic bodies and government agencies, standard clinical care pathways are available in the form of standard ontologies. Standard disease ontologies cover possible treatment options for patient-specific conditions. In this work, we have designed a framework to automatically segment kidney cancer/tumor to estimate the T staging and convert the T stage and other patient-specific clinical information to a standard Resource Definition Format (RDF). Converting patient-specific information in the form of ontologies will help to map with the standard ontologies to extract clinical care pathways. For kidney tumor segmentation in CT images, an unet-based deep learning architecture was used, and hyper parameters were tuned to get the tumor segmentation. Based on the size of the tumor, T-stage information was derived, and from patient electronic medical records, patient-specific clinical information was extracted. Using T-stage and clinical information, patient-specific clinical information was converted to RDF file format for each patient and stored in a Graph DB database that can be queried via Spark QL queries. To extract clinical care pathways for a specific condition of the patient, owl language was used to develop mapping functions for mapping patient-specific ontologies with standard ontologies. Proposed methodologies were tested on 209 patients, and we were able to achieve kidney segmentation and tumor segmentation accuracy of 0.9 and 0.86 IOU scores, respectively. Further owl functions were used to extract the clinical care pathways by mapping patient-specific ontologies to standard ontologies