AUTHOR=Ye Wei , Chen Xicheng , Li Pengpeng , Tao Yongjun , Wang Zhenyan , Gao Chengcheng , Cheng Jian , Li Fang , Yi Dali , Wei Zeliang , Yi Dong , Wu Yazhou 

TITLE=OEDL: an optimized ensemble deep learning method for the prediction of acute ischemic stroke prognoses using union features

JOURNAL=Frontiers in Neurology

VOLUME=Volume 14 - 2023

YEAR=2023

URL=https://www.frontiersin.org/journals/neurology/articles/10.3389/fneur.2023.1158555

DOI=10.3389/fneur.2023.1158555

ISSN=1664-2295

ABSTRACT=Background: Early stroke prognosis assessments are critical for decision-making regarding therapeutic interventions. We introduced the concepts of data combination, method integration, and algorithm parallelization, aiming to build an integrated deep learning model based on a combination of clinical and radiomics features and analyze its application value in prognosis prediction.
Methods: Utilizing data obtained from 441 stroke patients, clinical and radiomic features were extracted, and feature selection was performed. We used the concept of a deep ensemble to jointly analyze multiple deep learning methods, used metaheuristic algorithms to improve the efficiency of the parameter search process, and finally developed the optimized ensemble deep learning (OEDL) method. Finally, we evaluated and compared the predictive performance of each method.
Results: The combined features had better classification performance than the clinical and radiomic features, and the obtained accuracies (ACC) were 0.9043, 0.9321 and 0.9574 for the clinical, radiomics and combined models, respectively. Comparing the classification effects of the different methods, the results showed that compared with other single and ensemble methods, the OEDL method constructed based on the ensemble optimization concept could achieve the best classification performance, its ACC, Macro-PRE, Macro-SEN, Macro-F1, Macro-AUC reached 0.9574, 0.9403, 0.9475, 0.9435, 0.9789 respectively.
Conclusion: The OEDL approach proposed herein could effectively achieve improved stroke prognosis prediction performance, the effect of using combined data modeling was significantly better than that of single clinical or radiomics feature models, and the proposed method had better intervention guidance value. Our approach is beneficial for optimizing the early clinical intervention process and providing the necessary clinical decision support for personalized treatment.