Machine learning applications to clinical decision support in neurosurgery: an artificial intelligence augmented...

Area of Science:

• Neurosurgery
• Artificial Intelligence
• Data Science

Background:

• Machine learning (ML) algorithms analyze complex datasets for prediction and classification.
• ML holds potential for significant advancements in neurosurgical applications.
• Existing research on ML in neurosurgery is diverse, necessitating a systematic review.

Purpose of the Study:

• To systematically review current machine learning applications in neurosurgery.
• To assess the performance of various ML algorithms, including neural networks (NN), logistic regression (LR), and support vector machines (SVM).
• To identify research gaps and future opportunities for ML in neurosurgery.

Main Methods:

• Systematic literature search yielding 70 studies meeting inclusion criteria from 6866 results.
• Analysis of performance metrics such as area under the receiver operating characteristics curve (AUC), accuracy, sensitivity, and specificity.
• Application of natural language processing (NLP) for topic modeling and keyword identification within the neurosurgical corpus.

Main Results:

• The densest application of ML was in preoperative evaluation, planning, and outcome prediction, particularly in spine surgery.
• Significant differences in accuracy and specificity were observed among NN, LR, and SVM algorithms.
• NN algorithms demonstrated superior accuracy compared to LR, while SVM showed higher specificity than LR. No significant differences were found in AUC or sensitivity.
• NLP identified seven key topics related to modeling approach, surgery type, and pathology.

Conclusions:

• Machine learning technology accurately predicts outcomes and aids clinical decision-making in neurosurgery.
• Neural networks frequently outperform other algorithms in supervised learning tasks within neurosurgery.
• This review highlights opportunities for future research in neurosurgical ML applications.