Preparation of ultrasound contrast agents: The exploration of the structure-echogenicity relationship of contrast agents based on neural network model

There is a need to standardize the process of micro/nanobubble preparation to bring it closer to clinical translation. We explored a neural network-based model to predict the structure-echogenicity relationship for the preparation and fabrication of ultrasound-enhanced contrast agents. Seven formulations were screened, and 109 measurements were obtained. An artificial neural network-multilayer perceptron (ANN-MLP) model was used. The original data were divided into the training and testing groups, which included 73 and 36 groups of data, respectively. The hidden layer was selected from three hidden layers and included bias. The classification graph showed that the predicted values of the training and testing groups were 76.7% and 66.7%, respectively. According to the receiver operating characteristic curve, the accuracy of different imaging effects could achieve a prediction rate of 88.1–96.5%. The percentage graph showed that the data were gradually converging. The predictive analysis curves of different ultrasound effects gradually approached stable value of Gain. Normalized importance predicted contributions for the Pk1, poly-dispersity index (PDI), and intensity account were 100%, 98.5%, and 89.7%, respectively. The application of the ANN-MLP model is feasible and effective for the exploration of the synthesis process of ultrasound contrast agents. 1,2-Distearoyl-sn-glycero-3 phosphoethanolamine-N (methoxy[polyethylene glycol]-2000) (DSPE PEG-2000) correlated highly with the success rate of contrast agent synthesis.