An In-Depth Study of Personalized Anesthesia Management Models in Gastrointestinal Endoscopy Based on Multimodal Deep Learning

Abstract

In response to the annual occurrence of over 10 million gastrointestinal endoscopic examinations in China, this study proposes a personalized anesthesia management model based on multimodal deep learning. This model was designed to enhance anesthesia management efficiency and disease detection rates. In collaboration with the Department of Anesthesiology at Renji Hospital, which is affiliated with the Shanghai Jiao Tong University School of Medicine, data pertaining to anesthesia were collected from 398 patients, who were undergoing gastrointestinal endoscopy. This yielded a total of 327 valid samples. Analysis of the patients’ basic information and physiological parameters during surgery revealed that body mass index (BMI) and age significantly impacted anesthesia management. Based on these findings, a multimodal deep learning model was developed that integrates Long Short-Term Memory (LSTM) networks, hyperparameter geometric manifold optimization (GMO) methods and data-driven sparse matrix classifiers. The model is capable of dynamically adjusting its parameters based on the specific needs of each individual patient, utilizing real-time physiological data to predict vital signs and anesthesia states with a 10-second lead time. In the experimental evaluations, the model demonstrated superior performance in drug usage prediction tasks. Compared with LSTM networks integrated with convolutional neural networks (CNN) and support vector machines (SVM), the LSTM model combined with GMO and sparse matrix classifiers, along with personalized physiological data, achieved a recall rate of 83% and an F1-score of 0.711 in drug usage prediction. The total computation time was maintained within 2.99 seconds, thereby satisfying the requisite for real-time applications. This model significantly improves prediction accuracy and stability over traditional methods, thereby enhancing operational efficiency in complex surgical environments. It is anticipated that the outcomes of this study will promote widespread adoption of gastrointestinal endoscopy, thereby improving the early diagnosis and treatment rates of gastrointestinal diseases.

Keywords

• Gastrointestinal endoscopy
• LSTM networks
• hyperparameter optimization
• sparse matrix classifier
• real-time prediction