The research paper on AI Predictive Diagnosis Algorithm for Stroke (First author: Yosuke Hayashi, Responsible author: Taka-aki Nakada) has been published in the international scientific journal Scientific Reports (published by Nature Research, UK).
The research and development of the AI prediction algorithm for stroke was adopted by the Japan Agency for Medical Development (AMED) under the research and development project "Advanced Medical Devices and Systems Technology Development Project: Research and Development of Emergency Medicine Prediction", and was conducted jointly by Smart119 Inc. and the Department of Emergency and Intensive Care Medicine, Graduate School of Medicine, Chiba University.
This paper reports on the establishment and demonstrated effectiveness of an AI predictive diagnosis algorithm for stroke disease in acute care. This development is expected to be applied to other medical conditions. Smart119 Inc. has applied for a patent for this algorithm.
◆Detailed press release
https://prtimes.jp/main/html/rd/p/000000047.000056624.html
◆Background of the Development Research
The three major diseases ("cancer," "acute myocardial infarction," and "stroke") account for a large percentage of the number of emergency cases brought in.
Stroke, in particular, tends to occur suddenly and is classified into a number of conditions, including subarachnoid hemorrhage, cerebral infarction, cerebral hemorrhage, and occlusion of the main artery. In order to save lives as well as to reduce the aftereffects such as hemiplegia, prompt and optimal acute treatment is required. However, at present, the decisions of the emergency team are not shared with the medical institutions, so the medical condition of the patient is determined by diagnosis after arrival at the receiving hospital.
-Improving the accuracy of the emergency team's judgment based on the condition of the patient.
-Acute stage treatment at medical institutions with specialized doctors and facilities.
We have started to develop AI predictive diagnosis to satisfy the above two points quickly and accurately. The results of the AI predictive diagnosis will be shared between the emergency team and medical institutions.
◆Published in the international scientific journal Scientific Reports
A prehospital diagnostic algorithm for strokes using machine learning: a prospective observational study
https://www.nature.com/articles/s41598-021-99828-2
◆Purpose of Development
＜Purpose of the prediction algorithm＞
To determine the symptoms of stroke, such as "subarachnoid hemorrhage," "cerebral infarction," "cerebral hemorrhage," and "occlusion of the main artery," based on the conditions that exist individually in the background of emergency patients (condition, disease history, weather conditions, etc.).
＜Data formation＞
-Data will be collected from August 2018 with the cooperation of medical institutions in Chiba City and the Chiba City Fire Department.
-The number of data collected is approximately 1,500 emergency patients who may have had a stroke.
-The content of the data includes the condition of the emergency patient, age, gender, and weather conditions at the time.
＜Algorithm calculation method and verification＞
-XG Boost (*1)
-From the data of approximately 1,500 patients, design a classification algorithm and model using 80% (approximately 1,200 patients) for machine learning, and verify it using 20% (approximately 300 patients) for testing.
-As a result of validating the classification algorithm with the data of about 300 people for testing, an AUC value of 0.980(*2) was obtained.
◆Toward practical use
We expect that the AI predictive diagnosis function using the stroke prediction diagnosis algorithm will be added to the application "Smart119," an emergency medical information service, by the end of this year for tablet terminals installed in ambulances owned by the Chiba City Fire Department, which has introduced Smart119.
＜Procedure of AI predictive diagnosis function＞
1) If there is a possibility of stroke, tap the "Stroke Diagnosis Button".
2) On the dedicated diagnosis page, enter the emergency patient's condition according to the options.
3) Check the medical condition from the AI predictive diagnosis.
4) The system automatically selects a medical institution with specialized doctors and facilities, and makes a request for acceptance.
◆Published in the international scientific journal Scientific Reports
A prehospital diagnostic algorithm for strokes using machine learning: a prospective observational study
https://www.nature.com/articles/s41598-021-99828-2
◆Purpose of Development
＜Purpose of the prediction algorithm＞
To determine the symptoms of stroke, such as "subarachnoid hemorrhage," "cerebral infarction," "cerebral hemorrhage," and "occlusion of the main artery," based on the conditions that exist individually in the background of emergency patients (condition, disease history, weather conditions, etc.).
＜Data formation＞
-Data will be collected from August 2018 with the cooperation of medical institutions in Chiba City and the Chiba City Fire Department.
-The number of data collected is approximately 1,500 emergency patients who may have had a stroke.
-The content of the data includes the condition of the emergency patient, age, gender, and weather conditions at the time.
＜Algorithm calculation method and verification＞
-XG Boost (*1)
-From the data of approximately 1,500 patients, design a classification algorithm and model using 80% (approximately 1,200 patients) for machine learning, and verify it using 20% (approximately 300 patients) for testing.
-As a result of validating the classification algorithm with the data of about 300 people for testing, an AUC value of 0.980(*2) was obtained.
◆Toward practical use
We expect that the AI predictive diagnosis function using the stroke prediction diagnosis algorithm will be added to the application "Smart119," an emergency medical information service, by the end of this year for tablet terminals installed in ambulances owned by the Chiba City Fire Department, which has introduced Smart119.
＜Procedure of AI predictive diagnosis function＞
1) If there is a possibility of stroke, tap the "Stroke Diagnosis Button".
2) On the dedicated diagnosis page, enter the emergency patient's condition according to the options.
3) Check the medical condition from the AI predictive diagnosis.
4) The system automatically selects a medical institution with specialized doctors and facilities, and makes a request for acceptance.
＜ AI Predictive Diagnosis Screen＞
*This screen is a demo version and is possibly subject to change in the official release.
Using the tablet device, AI diagnosis can be performed and a request can be made to the most appropriate medical institution. Depending on the medical condition and symptoms, the receiving medical institution can call a specialist or prepare for emergency surgery before the ambulance arrives.
*1: A Scalable Tree Boosting System "Tree" is a thought diagram called a decision tree. A tree is a thought diagram in which branches spread out like a tree according to the branching points (decision points) in the process of reaching a goal. XG boosting is a method of improving accuracy by drawing a decision tree from the analysis of each model and using information on the differences between models.
*2：Area under the curve It is a curve value that indicates the accuracy of the classification algorithm. It is considered highly accurate when it exceeds the threshold value of 0.8.