Prediction model of acute exacerbation of chronic obstructive pulmonary disease based on machine learning / 中国康复理论与实践

ObjectiveIn view of the problems of large errors and poor accuracy in pulmonary function testing in patients with acute exacerbation of chronic obstructive pulmonary disease (AECOPD), a predictive classification model of pulmonary function in patients with AECOPD was proposed by comparing the prediction performance of different machine learning models to find the optimal model. MethodsFrom January, 2018 to February, 2020, 90 patients with different degrees of COPD from the Affiliated Suzhou Science and Technology Town Hospital of Nanjing Medical University were collected. Six machine learning model algorithms (K-nearest neighbor, logistic regression, support vector machine, naive Bayes, decision tree and random forest) were used to establish AECOPD predictive classification models. Their area under the curve of receiver operating characteristic (AUC-ROC) and accuracy were compared. Ten-fold cross-validation method was used to validate the data set. ResultsThe model based on random forest worked best in predicting and classifying AECOPD patients, with an accuracy rate of 0.844 and an AUC-ROC of 0.916. ConclusionRandom forest-based predictive model is a powerful tool for identifying patients with AECOPD, providing decision support when it is difficult to give a definitive diagnosis.

Quantitative assessment of motor function in patients with Parkinson's disease using wearable sensors / 生物医学工程学杂志

Motor dysfunction is the main clinical symptom and diagnosis basis of patients with Parkinson's disease (PD). A total of 30 subjects were recruited in this study, including 15 PD patients (PD group) and 15 healthy subjects (control group). Then 5 wearable inertial sensor nodes were worn on the bilateral upper limbs, lower limbs and waist of subjects. When completing the 6 paradigm tasks, the acceleration and angular velocity signals from different parts of the body were acquired and analyzed to obtain 20 quantitative parameters which contain information about the amplitude, frequency, and fatigue degree of movements to assess the motor function. The clinical data of the two groups were statistically analyzed and compared, and then Back Propagation (BP) Neural Network was used to classify the two groups and predict the clinical score. The final results showed that most of the parameters had significant difference between the two groups, ten times of 5-fold cross validation showed that the classification accuracy of the BP Neural Network for the two groups was 90%, and the predictive accuracy of Hoehn-Yahr (H-Y) staging and unified PD rating scale (UPDRS) Ⅲ score of the patients were 72.80% and 68.64%, respectively. This study shows the feasibility of quantitative assessment of motor function in PD patients using wearable sensors, and the quantitative parameters obtained in this paper may have reference value for future related research.