Original ArticleMachine Learning Models can Detect Aneurysm Rupture and Identify Clinical Features Associated with Rupture
Introduction
Existing tools to estimate rupture risk for unruptured intracranial aneurysms are limited. The International Study of Unruptured Intracranial Aneurysms (ISUIA)1 and other natural history studies2, 3, 4, 5 have consistently implicated size and location. The role of other clinical risk factors (e.g., smoking,6 multiple aneurysms,4 age7) remains unclear, and even our understanding of the importance of size4, 5 and location8, 9 continues to evolve. Recent attempts to understand rupture risk have taken new approaches, such as using hemodynamic modeling, morphologic analysis, and wall inflammation detection.10, 11, 12, 13, 14 Despite numerous studies seeking to better characterize the clinical and angiographic risk factors of aneurysmal rupture, our understanding remains limited.
New tools for analyzing large datasets may offer a modern approach to understanding aneurysm rupture risk. Machine learning (ML) is a type of artificial intelligence that can detect associations between features of a dataset without being explicitly programmed.15, 16, 17 Unlike classical statistics that require a hypothesis-driven approach to answering clinical questions, the statistical learning in ML allows for the discovery of unanticipated associations. ML algorithms not only detect important relationships but can be simply and rapidly applied to new data to make predictions.17 ML has been increasingly used throughout medicine, including neurosurgery, for diagnosis, surgical planning, and outcome prediction.15, 17, 18
Using 16 years of retrospective, single-institution data, we trained 3 unique ML models to classify rupture status and identify the clinical features most strongly associated with rupture based on basic patient data.
Section snippets
Patient Population
This retrospective study was approved by the Partners institutional review board (#2015P002352). Patient data were retrospectively reviewed for patients with unruptured or ruptured intracranial aneurysms detected on vascular imaging at Brigham and Women's Hospital between 2002 and 2018. Complete data were available for 615 patients in whom 845 aneurysms were found. Clinical features were extracted from admission notes, operative reports, and discharge documentation; radiographic data were
Results
A total of 845 aneurysms in 615 patients were included in the study. Of the 845 aneurysms, 309 (37%) were ruptured. A total of 473 patients had a single aneurysm; 142 patients had multiple aneurysms. A summary of the patient and aneurysm characteristics for the unruptured and ruptured cohorts is provided in Table 2. Ruptured aneurysms were larger (mean 6.51 mm vs. 5.73 mm; P = 0.02) and more likely to be in the posterior circulation (20% vs. 11%; P < 0.001) (Table 2).
Aneurysm location varied
Discussion
The importance of aneurysm size and location when estimating rupture risk is well-established. Our models corroborate the findings of ISUIA, unruptured cerebral aneurysms, and other natural history studies that have implicated size and location as the most significant risk factors for rupture. Moreover, the locations that our linear SVM model identified to be most positively (PCOMM, ACOMM) and negatively (paraclinoid, MCA) associated with rupture are consistent with what prior studies have
Conclusions
The ability of ML to identify complex relationships in large datasets makes it uniquely suited to the task of aneurysm rupture classification. The models developed in this study can accurately classify rupture status using predictors that are in accordance with prior literature. Our findings suggest that the ML techniques described here are reliable and have the potential for applications throughout neurosurgery.
References (24)
- et al.
Unruptured intracranial aneurysms: natural history, clinical outcome, and risks of surgical and endovascular treatment
Lancet
(2003) - et al.
Machine learning and neurosurgical outcome prediction: a systematic review
World Neurosurg
(2018) - et al.
The natural course of unruptured cerebral aneurysms in a Japanese cohort
N Engl J Med
(2012) - et al.
Natural history of unruptured intracranial aneurysms: probability of and risk factors for aneurysm rupture
J Neurosurg
(2000) - et al.
Small unruptured intracranial aneurysm verification study: SUAVe study, Japan
Stroke
(2010) - et al.
Lifelong rupture risk of intracranial aneurysms depends on risk factors: a prospective Finnish cohort study
Stroke
(2014) - et al.
Cigarette smoking and outcomes after aneurysmal subarachnoid hemorrhage: a nationwide analysis
J Neurosurg
(2018) - et al.
Natural history of unruptured intracranial aneurysms: a long-term follow-up study
Stroke
(2013) - et al.
Angioarchitectures and hemodynamic characteristics of posterior communicating artery aneurysms and their association with rupture status
AJNR Am J Neuroradiol
(2017) - et al.
Hemodynamic characteristics of ruptured and unruptured multiple aneurysms at mirror and ipsilateral locations
AJNR Am J Neuroradiol
(2017)
Association of hemodynamic factors with intracranial aneurysm formation and rupture: systematic review and meta-analysis
Neurosurgery
Prerupture intracranial aneurysm morphology in predicting risk of rupture: a matched case-control study
Neurosurgery
Cited by (42)
Patient-specific computational modelling of endovascular treatment for intracranial aneurysms
2023, Brain MultiphysicsArtificial intelligence and machine learning in aneurysmal subarachnoid hemorrhage: Future promises, perils, and practicalities
2023, Journal of the Neurological SciencesA framework for intracranial aneurysm detection and rupture analysis on DSA
2023, Journal of Clinical Neuroscience
Conflict of interest statement: The authors declare that the article content was composed in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.