High-performance prediction of epilepsy surgical outcomes based on the genetic neural networks and hybrid iEEG marker

Accurately identification of the seizure onset zone (SOZ) is pivotal for successful surgery in patients with medically refractory epilepsy. The purpose of this study is to improve the performance of model predicting the epilepsy surgery outcomes using genetic neural network (GNN) model based on a hybrid intracranial electroencephalography (iEEG) marker. We extracted 21 SOZ related markers based on iEEG data from 79 epilepsy patients. The least absolute shrinkage and selection operator (LASSO) regression was employed to integrated seven markers, selected after testing in pairs with all 21 biomarkers and 7 machine learning models, into a hybrid marker. Based on the hybrid marker, we devised a GNN model and compared its predictive performance for surgical outcomes with six other mainstream machine-learning models. Compared to the mainstream models, underpinning the GNN with the hybrid iEEG marker resulted in a better prediction of surgical outcomes, showing a significant increase of the prediction accuracy from approximately 87% to 94.3% (P = 0.0412). This study suggests that the hybrid iEEG marker can improve the performance of model predicting the epilepsy surgical outcomes, and validates the effectiveness of the GNN in characterizing and analyzing complex relationships between clinical data variables.

Simultaneous optimization of the extraction process of Yangyin Yiqi Huoxue prescription with natural deep eutectic solvents for optimal extraction yield and antioxidant activity: A comparative study of two models

INTRODUCTION

Natural deep eutectic solvents (NaDESs) are green and effective solvents that are used to extract 3 flavonoids from Yangyin Yiqi Huoxue prescription, a traditional Chinese prescription.

METHODS

A total of 6 types of NaDESs were systematically screened and evaluated for the total extraction yield of puerarin, calycosin, and formononetin by high-performance liquid chromatography. Then, a 4-factor-three-level experimental scheme designed by the Box-Benhnken Design was applied on the basis of a single experiment to determine the extraction yield and the antioxidant property. Finally, the extraction process was optimized through response surface methodology (RSM) and the genetic neural network (GNN), respectively.

RESULTS

The use of betaine-lactic acid as an extractant displayed significant advantages in the screening process. The optimum extraction parameters provided by GNN were as follows: water content 25% (v/v), liquid to material ratio 190 mg/ml, extraction time 37 min, and extraction temperature 63 °C. Under this condition, the average experimental comprehensive evaluation values of the extraction yield and antioxidant properties were 3.12 mg/g and 86.27%, and the relative deviations to the predicted values were 0.30% and 1.44%, respectively. In addition, the experimental results of GNN were better than those of RSM (p < 0.01).

CONCLUSIONS

We found the application of GNN to be effective and credible for bi-objective optimization of extraction yields and antioxidant activity in this study. Moreover, our results provide a reference and a theoretical basis for experimental and future industrial extraction for multi-objective situations.