ELECTRONICS AND TELECOMMUNICATIONS RESEARCH INSTITUTE;한국전자통신연구원
发明人:
정호열,최재훈,HO YOUL JUNG,JAE HUN CHOI
申请号：
KR1020190139631
公开号：
KR1020200069217A
申请日:
2019.11.04
申请国别(地区):
KR
年份:
2020
代理人:
摘要：
The apparatus for predicting the development of cardiovascular disease using heterogeneous data according to the present invention receives cardiac computed tomography image learning data from a first database and cardiac computed tomography image data of a patient to be predicted, and transmits to a plurality of patients from a second database. The first pre-learning weight is generated based on the first machine learning method using the network interface, cardiac computed tomography image learning data, which receives the time-series generated electronic medical record learning data and the predicted patient electronic medical record data, , Generates a second pre-learning weight based on the electronic compulsory record learning data based on the second machine learning method, and performs multi-modal machine learning based on the first pre-learning weight and the second pre-learning weight to generate a multi-modal prediction model. Based on the multi-modal prediction model learning department to build, based on the multi-modal prediction model, predictive outcome data on whether or not the future cardiovascular disease will occur is based on the predicted patient's cardiac computed tomography image data and the predicted patient's electronic medical record data. It includes an onset prediction unit to generate, a network interface, a multimodal prediction model learning unit, and a processor to control the onset prediction unit, but the electronic medical record learning data and the electronic medical record data of the patient to be predicted include Troponin I and Troponin T values. do.본 발명에 따른 이종 데이터를 이용하여 심혈관 질환 발병 예측 장치는 제1 데이터베이스로부터 심장 컴퓨터 단층 촬영 영상 학습 데이터 및 예측대상 환자의 심장 컴퓨터 단층 촬영 영상 데이터를 수신하고, 제2 데이터베이스로부터 복수의 환자들에 대하여 시계열적으로 생성된 전자 의무 기록 학습 데이터 및 예측 대상 환자의 전자 의무 기록 데이터를 수신하는 네트워크 인터페이스, 심장 컴퓨터 단층 촬영 영상 학습 데이터를 제1 기계 학습 방식에 기반하여 제1 사전학습 가중치를 생성하고, 전자 의무 기록 학습 데이터를 제2 기계 학습 방식에 기반하여 제2 사전학습 가중치를 생성하고, 제1 사전학습 가중치 및 제2 사전학습 가중치를 기반으로 멀티모달 기계 학습을 수행하여 멀티모달
