At Shizuoka University, the team of Yoshinobu Kano's laboratory in the Department of Behavioral Information, Faculty of Informatics announced that they have won the top grade in an international contest-type workshop to guess the diagnosis name from electronic medical records.
Currently, with the development of the artificial intelligence field expected, natural language processing, which automatically analyzes sentences, is one of the most important fields, and the increase in medical costs has become a national issue. Application to the field is considered to have great demand and social significance.
At the international contest-type workshop "NTCIR", the "MedNLP Task Series" will be held for several years to compete for medical language information processing in Japanese, and research teams from each country will compete for their performance to develop better systems. The purpose is.
In the "MedNLP Task Series", we have been competing for tasks such as the development of basic technologies such as extraction of disease names and drug names. This third task uses a simulated electronic medical record in Japanese and the international standard code "ICD-3" that classifies and numbers the diagnosis name and disease name, and uses only the text of the electronic medical record to identify the diagnosis name. A system that automatically selects from international standard codes is built to compete for performance.
The team at Kano Lab achieved the top grade in the strictest evaluation criteria (SURE criteria) in this task.Due to the difficulty of the task itself, the absolute performance is not high yet, but it is considered to be the first step to proceed with the construction of an automatic medical diagnosis support system in the future.The results of this time were announced at the NTCIR-2016 conference (results presentation) held from June 6, 7, and the papers have also been published.
