In emergency departments in Japan, physicians prepare for patients using information from emergency medical technicians before their arrival. But those information were not recorded on previous electronical health record system. We constructed a database to record prearrival information and developed experimental prediction models to assess whether the patients need hospital admission or not. The prediction accuracy of the models was superior to that of human predictions. We focused on the cases which physicians correctly predicted the outcome, and the models had almost perfect concordance. Our database and prediction model may support clinical decision making in emergency department.

The digitalization of medical treatment has progressed and huge amounts of medical data are accumulating. Electronic medical data include structured numerical data and unstructured text data. The medical text analysis is expected to improve medical process and the clinical decision support. The present paper analyzes the words that appear in operation records to predict the two peaks and the long hospitalization by Support vector machine, and evaluated them by Feature selection. Three measures were proposed and the prediction performance for importance of the feature word was evaluated. Two measures obtained that less than 10 words resulted in the optimal prediction performance. Moreover, it was confirmed the effect of medical dictionary.

With the recent rise in popularity and size of social media, there is a growing need for systems that can extract useful information from this amount of data. We have been addressing an issue of detecting influenza epidemics using social media. Although previous methods relied mainly on frequencies of the influenza related words, they suffered from the noisy words. To deal with this problem, this study proposes deeper Natural Language Processing (NLP) approach that focuses on the each person. This paper discus the basic feasibility of the proposed approach basded on our experiences.

Toward a final goal to construct a medical diagnostic support system, as its pilot study, we attempt to build a question-answering program that automatically answers the medical licensing examination.

To facilitate the meaningful use of medical record and report, we propose a new concept that can enhance expressiveness of structured medical records and reports. A medical record or a report based on the proposed concept is composed of entries and links, while a conventional one is composed of predefined sections, such as "FINDINGS" and "IMPRESSION" or such as "Subjective", "Objective", "Assessment" and "Plan", containing narrative sentences, respectively. Each entry can be an image annotation or a unit of description such as findings of a lesion, a possible diagnosis, or a recommendation. Links express relationships between pairs of entries, such as causation and correspondence. Links are characteristic components of such a record or report, and play an important role to provide a perspective on it. The proposed concept was implemented in a prototype reporting system that can cooperate with a PACS viewer. Using the system, 27 radiologists interpreted 54 CT studies including 27 follow-up ones. After the interpretation, they responded to a questionnaire on the system. According to the response, most of the radiologists thought that the prototype system was superior to conventional ones in many aspects. We believe, the proposed concept will be implemented in many hospital information systems, then play a pivotal role in linking various kinds of medical data among different departments to facilitate big-data utilization.

We have started to consider using a humanoid robot to the interface between the patients and the artificial intelligence component technologies in the future. We have tried to set several types of the verbal and non-verbal communication conditions of the robot. In this method, the user interface including the user experience might be able to provide to the patients. We believe that the user interface between the patients and the artificial intelligence component technologies might be able to manage using the humanoid robot.

医療情報は、診療目的としての利用、医療行為の公的書類作成のための利用（診療報酬請求）という一次利用だけでなく、社会的利用、医療政策立案・検証への利用、医学（疫学）研究への利用等の二次利用が考えられる。科学的根拠に基づいて安全かつ有効性に基づいた医療が患者に対して公平にいきわたるように、必要となる情報を収集し、還元すべきものと考えられる。医療情報は、カルテに記載される専門性の高い情報から、保険請求に関わるレセプト情報、個人で管理可能な健康診断、服薬情報、個人レベルの健康ライフログに至るまで、多種多様である。医療機関においても、大学病院といった専門性の高い医療を提供できる医療機関から、個人経営の診療所に至るまで扱う情報の質は、千差万別である。本来的にはそれらの情報は個人に帰属するものであるが、個人で管理しえないという特徴をもつ。こうした医療情報の特殊性は、医療提供体制の社会システムに関わる。 平成27年9月3日に「行政手続における特定の個人を識別するための番号の利用等に関する法律の一部改正）により、医療分野への適用することにより二次利用に向けた大きな期待と、同時に、医療情報を適切に管理し生かしていく上での課題が浮き彫りになった。課題には、今後の医療モデルを構築する上での技術的、社会的課題があり、以下に例示する。以下に、例示する。 ・前提としてのデータの標準化と相互運用性の確保。 ・前提としての医療機関間の相互連携のシステム化と医療機関の個人への医療情報開示の促進。 ・匿名化技術の利用等によるプライバシー保護の徹底。 ・受益者にオープンデータ化した医療情報を開示し、多様な視点からの分析を促進するとともに、受益者のコンセンサスに基づく、医療サービスの充実とコストの最適化を追求

Recently, machine learning techniques have spread many fields. However, machine learning is still not popular in medical research field due to difficulty of interpreting. In this paper, we introduce a method of interpreting medical information using machine learning technique. The method gave new explanation of partial dependence plot from medical research field.

Most of the severity ratings are assessed through interview with patients in psychiatric filed. Such severity ratings sometimes lack objectivity that can lead to the delay/misjudgment of the treatment initiation/switch. A new technology which enables us to objectively quantify patients' severity is needed. We here aim to develop a new device that analyzes patients' facial expression, voice, and daily activities, and provides us with objective severity evaluation using machine learning technology. This study project was accepted by Japan Agency for Medical Research and Development (AMED) and will launch this year. The background of the study purpose and methods will be presented.

In this paper we briefly review time-series medical data mining techniques and discuss problems toward knowledge discovery from heterogeneous time-series medical data composed of labo exams, prescriptions and descriptions with narratives.