New morphological parameter for intracranial aneurysms and rupture risk prediction based on artificial neural networks

Machine Learning Algorithms to Predict the Risk of Rupture of Intracranial Aneurysms: a Systematic Review

PURPOSE

Subarachnoid haemorrhage is a potentially fatal consequence of intracranial aneurysm rupture, however, it is difficult to predict if aneurysms will rupture. Prophylactic treatment of an intracranial aneurysm also involves risk, hence identifying rupture-prone aneurysms is of substantial clinical importance. This systematic review aims to evaluate the performance of machine learning algorithms for predicting intracranial aneurysm rupture risk.

METHODS

MEDLINE, Embase, Cochrane Library and Web of Science were searched until December 2023. Studies incorporating any machine learning algorithm to predict the risk of rupture of an intracranial aneurysm were included. Risk of bias was assessed using the Prediction Model Risk of Bias Assessment Tool (PROBAST). PROSPERO registration: CRD42023452509.

RESULTS

Out of 10,307 records screened, 20 studies met the eligibility criteria for this review incorporating a total of 20,286 aneurysm cases. The machine learning models gave a 0.66-0.90 range for performance accuracy. The models were compared to current clinical standards in six studies and gave mixed results. Most studies posed high or unclear risks of bias and concerns for applicability, limiting the inferences that can be drawn from them. There was insufficient homogenous data for a meta-analysis.

CONCLUSIONS

Machine learning can be applied to predict the risk of rupture for intracranial aneurysms. However, the evidence does not comprehensively demonstrate superiority to existing practice, limiting its role as a clinical adjunct. Further prospective multicentre studies of recent machine learning tools are needed to prove clinical validation before they are implemented in the clinic.

Diagnostic and predictive value of radiomics-based machine learning for intracranial aneurysm rupture status: a systematic review and meta-analysis

Currently, the growing interest in radiomics within the clinical practice has prompted some researchers to differentiate the rupture status of intracranial aneurysm (IA) by developing radiomics-based machine learning models. However, systematic evidence supporting its performance remains scarce. The purpose of this meta-analysis and systematic review is to assess the diagnostic performance of radiomics-based machine learning for the early detection of IA rupture and to offer evidence-based recommendations for the application of radiomics in this area. PubMed, Cochrane, Embase, and Web of Science databases were searched systematically up to March 2, 2024. The Radiomics Quality Score (RQS) was employed to assess the risk of bias in all included primary studies. We separately discussed the diagnostic or predictive performance of machine learning for IA rupture status based on task type (diagnosis or prediction).  We finally included 15 original studies covering 9,111 IA cases. In the validation cohort, radiomics demonstrated a sensitivity, specificity, positive likelihood ratio, negative likelihood ratio, diagnostic odds ratio, as well as SROC curve of 0.84 (95% CI: 0.76-0.90), 0.82 (95% CI: 0.77-0.86), 4.7 (95% CI: 3.7-5.8), 0.19 (95% CI: 0.13-0.29), and 24 (95% CI: 15-40), respectively, for the diagnostic task of aneurysm rupture status. Only 2 studies (3 models) addressed predictive tasks, with sensitivity and specificity ranging from 0.77 to 0.89 and from 0.69 to 0.87, respectively. Radiomics-based machine learning exhibits promising accuracy for early identification of IA rupture status, whereas evidence for its predictive capability is limited. Further research is needed to validate predictive models and provide insights for developing specialized strategies to prevent aneurysm rupture.

Investigation of paraclinoid aneurysm formation by comparing the combined influence of hemodynamic parameters between aneurysmal and non-aneurysmal arteries

Numerous studies have evaluated the effects of hemodynamic parameters on aneurysm formation. However, the reasons why aneurysms do not initiate in intracranial arteries are still unclear. This study aimed to investigate the influence of hemodynamic parameters, wall shear stress (WSS) and strain, on aneurysm formation by comparing between aneurysmal and non-aneurysmal arteries. Fifty-eight patients with paraclinoid aneurysms on one side were enrolled. Based on magnetic resonance angiography, each patient's left and right internal carotid arteries (ICAs) were reconstructed. For a patient having an aneurysm on one side, the ICA with the paraclinoid aneurysm was defined as the aneurysmal artery after eliminating the aneurysm, whereas the opposite ICA without aneurysm was defined as the non-aneurysmal artery. Computational fluid dynamics and fluid-structure interaction analyses were then performed for both aneurysmal and non-aneurysmal arteries. Finally, the relationship between high hemodynamic parameters and aneurysm location was investigated. For aneurysmal arteries, high WSS and strain locations were well-matched with the aneurysm formation site. Also, considerable correlations between high WSS and strain locations were observed. However, there was no significant relationship between high hemodynamic parameters and aneurysm formation for non-aneurysmal arteries. The findings are helpful for understanding aneurysm formation mechanism and encouraging further relevant research.

Nomogram-based geometric and hemodynamic parameters for predicting the growth of small untreated intracranial aneurysms

Previous studies have shown that the growth status of intracranial aneurysms (IAs) predisposes to rupture. This study aimed to construct a nomogram for predicting the growth of small IAs based on geometric and hemodynamic parameters. We retrospectively collected the baseline and follow-up angiographic images (CTA/ MRA) of 96 small untreated saccular IAs, created patient-specific vascular models and performed computational fluid dynamics (CFD) simulations. Geometric and hemodynamic parameters were calculated. A stepwise Cox proportional hazards regression analysis was employed to construct a nomogram. IAs were stratified into low-, intermediate-, and high-risk groups based on the total points from the nomogram. Receiver operating characteristic (ROC) analysis, calibration curves, decision curve analysis (DCA) and Kaplan-Meier curves were evaluated for internal validation. In total, 30 untreated saccular IAs were grown (31.3%; 95%CI 21.8%-40.7%). The PHASES, ELAPSS, and UIATS performed poorly in distinguishing growth status. Hypertension (hazard ratio [HR] 4.26, 95%CI 1.61-11.28; P = 0.004), nonsphericity index (95%CI 4.10-25.26; P = 0.003), max relative residence time (HR 1.01, 95%CI 1.00-1.01; P = 0.032) were independently related to the growth status. A nomogram was constructed with the above predictors and achieved a satisfactory prediction in the validation cohort. The log-rank test showed significant discrimination among the Kaplan-Meier curves of different risk groups in the training and validation cohorts. A nomogram consisting of geometric and hemodynamic parameters presented an accurate prediction for the growth status of small IAs and achieved risk stratification. It showed higher predictive efficacy than the assessment tools.

Influence of circle of Willis modeling on hemodynamic parameters in anterior communicating artery aneurysms and recommendations for model selection

Computational fluid dynamics (CFD) has been utilized to calculate hemodynamic parameters in anterior communicating artery aneurysm (AComA), which is located at a junction between left and right A1 and A2 segments. However, complete or half circle of Willis (CoW) models are used indiscriminately. This study aims to suggest recommendations for determining suitable CoW model. Five patient-specific CoW models with AComA were used, and each model was divided into complete, left-half, and right-half models. After validating the CFD using a flow experiment, the hemodynamic parameters and flow patterns in five AComAs were compared. In four out of five cases, inflow from one A1 side had a dominant influence on the AComA, while both left and right A1 sides affected the AComA in the remaining case. Also, the average difference in time-averaged wall shear stress between the complete and half models for four cases was 4.6%, but it was 62% in the other case. The differences in the vascular resistances of left and right A1 and A2 segments greatly influenced the flow patterns in the AComA. These results may help to enhance clinicians' understanding of blood flow in the brain, leading to improvements in diagnosis and treatment of cerebral aneurysms.

Prediction of cerebral aneurysm rupture risk by machine learning algorithms: a systematic review and meta-analysis of 18,670 participants

It is possible to identify unruptured intracranial aneurysms (UIA) using machine learning (ML) algorithms, which can be a life-saving strategy, especially in high-risk populations. To better understand the importance and effectiveness of ML algorithms in practice, a systematic review and meta-analysis were conducted to predict cerebral aneurysm rupture risk. PubMed, Scopus, Web of Science, and Embase were searched without restrictions until March 20, 2023. Eligibility criteria included studies that used ML approaches in patients with cerebral aneurysms confirmed by DSA, CTA, or MRI. Out of 35 studies included, 33 were cohort, and 11 used digital subtraction angiography (DSA) as their reference imaging modality. Middle cerebral artery (MCA) and anterior cerebral artery (ACA) were the commonest locations of aneurysmal vascular involvement-51% and 40%, respectively. The aneurysm morphology was saccular in 48% of studies. Ten of 37 studies (27%) used deep learning techniques such as CNNs and ANNs. Meta-analysis was performed on 17 studies: sensitivity of 0.83 (95% confidence interval (CI), 0.77-0.88); specificity of 0.83 (95% CI, 0.75-0.88); positive DLR of 4.81 (95% CI, 3.29-7.02) and the negative DLR of 0.20 (95% CI, 0.14-0.29); a diagnostic score of 3.17 (95% CI, 2.55-3.78); odds ratio of 23.69 (95% CI, 12.75-44.01). ML algorithms can effectively predict the risk of rupture in cerebral aneurysms with good levels of accuracy, sensitivity, and specificity. However, further research is needed to enhance their diagnostic performance in predicting the rupture status of IA.

Influence of blood viscosity models and boundary conditions on the computation of hemodynamic parameters in cerebral aneurysms using computational fluid dynamics

BACKGROUND

Computational fluid dynamics (CFD) is widely used to calculate hemodynamic parameters that are known to influence cerebral aneurysms. However, the boundary conditions for CFD are chosen without any specific criteria. Our objective is to establish the recommendations for setting the analysis conditions for CFD analysis of the cerebral aneurysm.

METHOD

The plug and the Womersley flow were the inlet boundary conditions, and zero and pulsatile pressures were the outlet boundary conditions. In addition, the difference in the assumption of viscosity was analyzed with respect to the flow rate. The CFD process used in our research was validated using particle image velocimetry experiment data from Tupin et al.'s work to ensure the accuracy of the simulations.

RESULTS

It was confirmed that if the entrance length was sufficiently secured, the inlet and outlet boundary conditions did not affect the CFD results. In addition, it was observed that the difference in the hemodynamic parameter between Newtonian and non-Newtonian fluid decreased as the flow rate increased. Furthermore, it was confirmed that similar tendencies were evaluated when these recommendations were utilized in the patient-specific cerebral aneurysm models.

CONCLUSIONS

These results may help conduct standardized CFD analyses regardless of the research group.