Multinomial Classification of Neurosurgical Operations Using Gradient Boosting and Deep Learning Algorithms.

Area of Science:

• Neurosurgery
• Computer Science
• Medical Informatics

Background:

• Accurate classification of neurosurgical procedures is crucial for research, education, and administrative purposes.
• Manual classification is time-consuming and prone to errors.
• The potential of artificial intelligence for automating this task remains largely unexplored.

Purpose of the Study:

• To evaluate the feasibility of classifying over 100 neurosurgical procedures using natural language processing (NLP) and machine learning (ML).
• To assess the performance of different ML algorithms in this classification task.

Main Methods:

• Utilized a dataset of neurosurgical procedure descriptions.
• Applied NLP techniques to process the text data.
• Trained and evaluated machine learning models, including a CatBoost algorithm and a bidirectional recurrent neural network with gated recurrent units (GRU).

Main Results:

• Both CatBoost and bidirectional RNN-GRU models achieved comparable accuracies of approximately 81% for classifying neurosurgical procedures.
• The models demonstrated high performance in suggesting the correct procedure class within the top 2-3 predictions, reaching up to 98.9% recall.
• The study confirmed the technical feasibility of ML-based classification for a large number of neurosurgical procedures.

Conclusions:

• Machine learning, particularly using NLP, presents a technically viable solution for classifying neurosurgical procedures into numerous categories.
• Further improvements in accuracy and reliability can be achieved through data enhancement and rigorous class verification.
• This approach holds promise for streamlining documentation, improving data analysis, and advancing neurosurgical research.