School of Artificial Intelligence and Data Science, Hebei University of Technology, Tianjin, China; Engineering Research Center of Intelligent Rehabilitation Device and Detection Technology, Ministry of Education, Tianjin, China.
School of Artificial Intelligence and Data Science, Hebei University of Technology, Tianjin, China; Engineering Research Center of Intelligent Rehabilitation Device and Detection Technology, Ministry of Education, Tianjin, China; Qinhuangdao Institute of Rehabilitation Technical Aids, NRRA, Qinhuangdao, Hebei, China.
World Neurosurg. 2022 Feb;158:e662-e674. doi: 10.1016/j.wneu.2021.11.040. Epub 2021 Nov 15.
Because of the complex condition of patients with spinal cord injury (SCI), it is difficult to accurately calculate the activity of daily living (ADL) score of discharged patients. In view of the above problem, this research proposes a prediction model of discharged ADL score based on machine learning, in order to get the rehabilitation effect of patients after rehabilitation training.
First, the medical records of 1231 patients with SCI were collected, and the corresponding data preprocessing was carried out. Secondly, the Pearson correlation coefficient method was combined with the feature selection method based on random forest (RF) to screen out 6 features closely related to the discharged ADL score. Then RF and RF optimized by Harris hawks optimizer (HHO-RF) were used to predict the discharged ADL score of patients with SCI. The mean absolute error (MAE), root mean squared error (RMSE), and coefficient of determination (R) were used as evaluation indicators of the model.
The prediction features selected by feature extraction were ADL score on admission, age, injury segment, injury reason, injury position, and injury degree. After 10-fold cross-validation, MAE, RMSE, and R of RF were 0.0875, 0.1346, and 0.7662, respectively. MAE, RMSE, and R of HHO-RF were 0.0821, 0.1089, and 0.8537, respectively. The prediction effect of HHO-RF has been greatly improved.
In clinical treatment, HHO-RF can accurately predict the discharged ADL score and provide a reasonable direction for patients to choose rehabilitation programs.
由于脊髓损伤（SCI）患者的病情复杂，难以准确计算出院患者的日常生活活动（ADL）评分。针对上述问题，本研究提出了一种基于机器学习的出院 ADL 评分预测模型，以了解患者康复训练后的康复效果。
首先收集了 1231 名 SCI 患者的病历，并进行了相应的数据预处理。其次，结合 Pearson 相关系数法和基于随机森林（RF）的特征选择方法，筛选出与出院 ADL 评分密切相关的 6 个特征。然后使用 RF 和经哈里斯鹰优化器（HHO-RF）优化的 RF 预测 SCI 患者的出院 ADL 评分。采用平均绝对误差（MAE）、均方根误差（RMSE）和决定系数（R）作为模型的评价指标。
特征提取选择的预测特征为入院时的 ADL 评分、年龄、损伤节段、损伤原因、损伤部位和损伤程度。经过 10 折交叉验证，RF 的 MAE、RMSE 和 R 分别为 0.0875、0.1346 和 0.7662。HHO-RF 的 MAE、RMSE 和 R 分别为 0.0821、0.1089 和 0.8537。HHO-RF 的预测效果得到了很大的提高。
在临床治疗中，HHO-RF 可以准确预测出院 ADL 评分，为患者选择康复方案提供合理的方向。
