Date of Award

January 2024

Document Type

Open Access Thesis

Degree Name

Medical Doctor (MD)

Department

Medicine

First Advisor

Michael DiLuna

Abstract

INTRODUCTION: Artificial intelligence (AI) and machine learning (ML) techniques have become highly visible in society and medicine, with some AI/ML-enabled devices receiving approval by the Federal Drug Administration for use in direct patient care. In neurosurgery, ML algorithms have been developed for clinical outcome prediction, many of which have achieved higher predictive ability than standard statistical analysis techniques. However, few have synthesized the ML literature in neurosurgery and its subspecialties. Given the rapid rate at which ML algorithms for outcome prediction and prognostication have developed in neurosurgery, additional synthesis of the current ML landscape in this area is necessary. The aim of this study was to perform a rapid review of PubMed-indexed neurosurgery literature concerning the development, validation, and/or use of ML algorithms for clinical outcome prediction. METHODS: Studies describing the development, validation, and/or use of ML algorithms for clinical outcome prediction published through October 15, 2023 in 14 prominent neurosurgery journals were identified via PubMed. Articles were screened by title and abstract. Manuscripts passing title-abstract screening were manually reviewed for inclusion. Studies were placed into groups based on subspeciality. Studies concerning the use of ML for radiomics were excluded. RESULTS: A total of 741 articles were identified from the initial PubMed query. Of these articles, 247 articles (33.3%) passed title-abstract screening. Of the articles that passed initial screening, 202 were included (27.3% of original 741 articles; 81.8% of articles that passed title-abstract screening). Of the 202 articles that passed both initial title-abstract screening and manuscript review, 114 (56.4%) were in the Spine cohort, 29 (14.4%) were in the Neuro-Oncology cohort, 7 (3.5%) were in the Pediatric Neurosurgery cohort, 31 (15.3%) were in the Cerebrovascular Neurosurgery cohort, 4 (2.0%) were in the Epilepsy/Functional Neurosurgery cohort, 12 (5.9%) were in the Trauma cohort, and 5 (2.5%) were in the Other cohort. External validation was performed in 22 studies (10.9%). Many reported algorithms achieved high model performance for clinical outcome prediction in neurosurgery and its subspecialties. CONCLUSION: The ML literature for clinical outcome precision in neurosurgery has grown rapidly in recent years. Few articles report external validation of developed ML algorithms, limiting their generalizability and clinical practicality. Future ML algorithms developed for clinical outcome prediction in neurosurgery should perform external validation of developed ML models.

Comments

This is an Open Access Thesis.

Download
Open Access

This Article is Open Access

Share

COinS