Research on adverse event classification algorithm of da Vinci surgical robot based on Bert-BiLSTM model

Area of Science:

• Medical device safety
• Natural Language Processing
• Machine Learning in Healthcare

Background:

• Adverse event reporting for medical devices like the da Vinci surgical robot often suffers from incomplete and inconsistent data.
• Accurate classification of these events is crucial for ensuring patient safety and improving medical device performance.

Purpose of the Study:

• To enhance the accuracy of classifying adverse events associated with the da Vinci surgical robot.
• To develop a robust model for predicting patient harm from adverse event reports, particularly for small datasets.

Main Methods:

• Developed a novel deep learning model, Bert-BiLSTM-Att_dropout, combining BERT and BiLSTM with attention and dropout mechanisms.
• Applied the model to a dataset of 4,568 da Vinci surgical robot adverse event reports (2013-2023).
• Evaluated model performance against baseline models like GRU, LSTM, and BiLSTM-Attention.

Main Results:

• The Bert-BiLSTM-Att_dropout model achieved an average F1 score of 90.15%.
• Significantly outperformed traditional and attention-based baseline models in text classification accuracy.
• Demonstrated effective generalization and key information capture on a small dataset.

Conclusions:

• The proposed model significantly improves the accuracy and usability of adverse event reporting for medical devices.
• This advancement contributes to preventing medical incidents and reducing patient harm.
• The findings highlight the potential of combined BERT and BiLSTM models for medical text classification tasks.