Core 1: Knowledge Extraction
To extract knowledge from multiple resources including structured and unstructured, represent it in the form of Ripple Down Rules (RDR) which is easy to use and evolvable with minimum efforts.
The primary feature of Knowledge Extraction include extracting knowledge from divers resources and convert that into actionable/executable knowledge. To achieve this, we go through some of the other features described as follows.
- Accept structured data (EMR, EHR) and unstructured data (text, images) as input for knowledge extraction.
- To extract knowledge from textual resources, machine learning and NLP techniques are performed to generate n-grams, named entities (concepts), concept hierarchy (concept tree) and attaches them to the exiting domain ontology.
- The main features of knowledge extraction from medical images includes category classification, segmentation, local feature extraction and model collation.
- The knowledge extracted from textural resources in the form of ontology, and from visual resources are utilized by Actionable Knowledge component to transform the extracted knowledge to actionable/executable form.
- Actionable knowledge is stored in the form of RDR to verify and evolve with minim effort and with intervention of knowledge engineers.
Uniqueness & Contribution
- A novel approach of evolutionary knowledge acquisition and modeling with real time verification..
- Automated ontology construction and evolution.
- Enhancing image quality with noise reduction and contrast enhancement for knowledge extraction.
- Generate specific model for particular image type and organ types.
- Aims to develop clinical knowledge generation and semantic inference techniques that comply with medical standard formats for various types of medical orthopedic / informal data
- Development of non-character knowledge acquisition component
- Development of unstructured knowledge acquisition component
- Formal Knowledge Acquisition Component Development
- Development of semantic reasoning engine
- Achieve consistency and interoperability of knowledge using medical standard formats
- Achieve efficiency and accuracy by extracting hidden knowledge in medical data Extraction of clinical knowledge from non-character medical images based on deep learning
- Automatic selection of optimal inference engine for different domains and improvement of reasoning ability based on semantic rule
Prof. Sooyong Shin
- Machine learning, Medical information, Biomedical Informatics
Prof. Byeong Ho Kang
- AI, Expert system
Prof. Joon Beom Seo
- Thoracic imaging