Deep learning model integrating radiologic and clinical data to predict mortality after ischemic stroke
Changi Kim,
Joon-myoung Kwon,
Jiyeong Lee,
Hongju Jo,
Dowan Gwon,
Jae Hoon Jang,
Min Kyu Sung,
Sang Won Park,
Chulho Kim,
Mi-Young Oh
Affiliations
Changi Kim
Department of Bioengineering, Seoul National University, Seoul, Republic of Korea
Joon-myoung Kwon
Medical Research Team, Medical AI Inc, DC, USA; Department of Critical Care Emergency Medicine, Incheon Sejong Hospital, Incheon, Republic of Korea; Artificial Intelligence and Big Data Research Center, Sejong Medical Research Institute, Bucheon, Republic of Korea
Jiyeong Lee
Department of Neurology, Bucheon Sejong Hospital, Bucheon, Republic of Korea
Hongju Jo
Independent Researcher
Dowan Gwon
Department of Digital&Biohealth, Group of AI/DX Business, KT, Seoul, Republic of Korea
Jae Hoon Jang
Department of Family Medicine, College of Medicine, KyungHee University, Seoul, Republic of Korea
Min Kyu Sung
Department of Family Medicine, College of Medicine, KyungHee University, Seoul, Republic of Korea
Sang Won Park
Department of Medical Informatics, School of Medicine, Kangwon National University, Chuncheon, Republic of Korea; Institute of Medical Science, School of Medicine, Kangwon National University, Chuncheon, Republic of Korea
Chulho Kim
Department of Neurology, Hallym University College of Medicine, Chuncheon, Republic of Korea; Corresponding author. Department of Neurology, Hallym University College of Medicine, 1 Hallymdaehak-gil, Chuncheon-si, Gangwon-do, 24252, Republic of Korea.
Mi-Young Oh
Department of Neurology, Bucheon Sejong Hospital, Bucheon, Republic of Korea; Corresponding author. Department of Neurology, Bucheon Sejong Hospital, 28, Hohyeon-ro 489, beon-gil, Bucheon-si, Gyeonggi-do, 14754, Republic of Korea.
Abstracts: Objective: Most prognostic indexes for ischemic stroke mortality lack radiologic information. We aimed to create and validate a deep learning-based mortality prediction model using brain diffusion weighted imaging (DWI), apparent diffusion coefficient (ADC), and clinical factors. Methods: Data from patients with ischemic stroke who admitted to tertiary hospital during acute periods from 2013 to 2019 were collected and split into training (n = 1109), validation (n = 437), and internal test (n = 654). Data from patients from secondary cardiovascular center was used for external test set (n = 507). The algorithm for predicting mortality, based on DWI and ADC (DLP_DWI), was initially trained. Subsequently, important clinical factors were integrated into this model to create the integrated model (DLP_INTG). The performance of DLP_DWI and DLP_INTG was evaluated by using time-dependent area under the receiver operating characteristic curves (TD AUCs) and Harrell concordance index (C-index) at one-year mortality. Results: The TD AUC of DLP_DWI was 0.643 in internal test set, and 0.785 in the external dataset. DLP_INTG had a higher performance at predicting one-year mortality than premise score in internal dataset (TD- AUC: 0.859 vs. 0.746; p = 0.046), and in external dataset (TD- AUC: 0.876 vs. 0.808; p = 0.007). DLP_DWI and DLP_INTG exhibited strong discrimination for the high-risk group for one-year mortality. Interpretation: A deep learning model using brain DWI, ADC and the clinical factors was capable of predicting mortality in patients with ischemic stroke.
