Towards a Pancreatic Lesions Disease Classification System based on Ontologies

摘要

Pancreatic cysts may have a significant malignant potential and their identification represents a `window of opportunity' for the early detection of pancreatic cancer. Previous works have achieved certain accuracy for classification on segmented cystic lesions in CT. However, pancreatic cysts have a large variation in size and shape, and the precise results of its segmentation remains rather challenging, for which we propose a disease-diagnosis ontology for early differential diagnosis of pancreatic cysts without pre-segmenting the lesions using an ontology-based-classification. Disease-diagnosis ontology is being developed to model and represent diagnosis information found in medical abstracts. The ontology studies the high-level features of cysts and builds mappings between medical imaging appearances to different pathological types of pancreatic cysts. To improve clinical applicability, we integrate a classifier of each cyst based on the results of past patients. The test on a group of 73 patients' pancreatic lesion cystic pathologically confirmed, reached an overall accuracy of 86%, which is significantly higher than the previous accuracy of 51.4%, which strongly reinforces the clinical potential of our developed method. In this article, we present the architecture of the approach, the key processes, and the first steps of the experimental validation through the developed feasibility model.