基于Attention的多源多任务ICU疾病诊断模型
DMMAM : Deep Multi-Source Multi-Task Attention Model for Intensive Care Unit Diagnosis
DASFAA 2019: Database Systems for Advanced Applications
Apr 22, 2019 - Apr 25, 2019.
Chiang Mai, Thailand
Zhenkun Shi, Wanli Zuo, Weitong Chen, Lin Yue, Yuwei Hao, Shining Liang*.
Abstract
Disease diagnosis can provide crucial information for clin- ical decisions that influence the outcome in acute serious illness, and this is particularly in the intensive care unit (ICU). However, the cen- tral role of diagnosis in clinical practice is challenged by evidence that does not always benefit patients and that factors other than disease are important in determining patient outcome. To streamline the diagnos- tic process in daily routine and avoid misdiagnoses, in this paper, we proposed a deep multi-source multi-task attention model (DMMAM) for ICU disease diagnosis. DMMAM exploits multi-sources information from various types of complications, clinical measurements, and the medical treatments to support the diagnosis. We evaluate the proposed model with 50 diseases of 9 classifications on an extensive collection of real- world ICU Electronic Health Records (EHR) dataset with 151729 ICU admissions from 46520 patients. Experiments results demonstrate the effectiveness and the robustness of our model.
Attachment
DMMAM : Deep Multi-Source Multi-Task Attention Model for Intensive Care Unit Diagnosis