Table 1. Demonstrates a summary of the utilities of AI in the enhancement of AAA management according to the sources used.

AAA, abdominal aortic aneurysm; ML, machine learning; SVR, support vector regression; ANNs, artificial neural networks; BN, Bayesian networks; LOS, length of stay; PCDF, posterior cervical decompression with instrumented fusion; ARVA, augmented radiology for vascular aneurysm; HPSE, heparinase

Study	Year of publication	Method	AI model	Cohort	Results
Raffort, et al. [4]	2020	Data extraction was done, and titles and abstracts were assessed independently by two authors. 34 studies with distinct methods, goals, and research designs were found after a thorough review of the published literature.	Various AI models (not specified)	34 studies	In addition, several prognostic and predictive instruments were developed to assess patient outcomes after surgery, including death rates and complications after endovascular aneurysm repair.
Xiong, et al. [5]	2022	By employing various ML techniques to discern biomarkers distinguishing large AAA from small AAA. Validation of these biomarkers was conducted using GEO datasets. Employing CIBERSORT, evaluation of immune cell infiltration in AAA tissues alongside the exploration of correlation between biomarkers and infiltrating immune cells.	LASSO, SVM-RFE, and RF	288 differentially expressed genes	G0/G1 switch 2 (G0S2) showed strong discriminatory power as an AAA biomarker with AUC values of 0.861, 0.875, and 0.911 in GSE57691, GSE47472, and GSE7284, respectively. For large AAA, heparinase (HPSE) had AUC values of 0.669 and 0.754 in GSE57691 and GSE98278, respectively, confirmed by qRT-PCR.
Cabrera, et al. [6]	2023	The study employed the RF algorithm to analyze data from the ACS-NSQIP database spanning 2008 to 2018. It aimed to predict outcomes including LOS, readmission, reoperation, transfusion, and infection rates following elective PCDF. Independent clinical variables' significance in predicting these outcomes was evaluated using the reduction in the Gini index.	RF algorithm	12,913 patients	The ACS-NSQIP database analysis identified key patient characteristics and perioperative events for elective PCDF, such as post-operative infection, age, BMI, operative time, LOS, preoperative hematocrit, and white blood cell count. The study highlighted risk factors for reoperation, readmission, hospital LOS, transfusion needs, and post-operative infection, along with their respective AUC values.
Lee, et al. [8]	2018	To predict the future growth of AAA for individual patients, a benchmark ML method known as non-linear Kernel SVR is used. The approach taken used baseline FMD and AAA diameter as input variables.	Non-linear Kernel SVR	94 patients from OxAAA	Growth data were prospectively collected from 94 patients at 12 months and from 79 patients at 24 months. The average increase in AAA diameter was 3.4% at 12 months and 2.8% annually at 24 months. The ML algorithm accurately predicted individual AAA diameters within a 2 mm margin of error for 85% and 71% of patients at 12 and 24 months, respectively.
Karthikesaling-am, et al. [12]	2016	Aneurysm morphology was assessed pre-operatively, and endograft complications were monitored for up to 5 years post-surgery. Using ANN, researchers predicted patients' risk levels for endograft complications or mortality. Centre 1 data trained the ANN, and Centre 2 data validated it. ANN performance was evaluated using Kaplan-Meier analysis, comparing the occurrence of complications and mortality between predicted low-risk and high-risk patients.	ANN	761 patients	A total of 761 patients, with a mean age of 75 +/- 7 years, underwent EVAR, and were followed up for an average of 36+/- 20 months. The ANN, which integrated morphological features, effectively forecasted the risk of endograft complications and mortality. External validation revealed significant differences in the five-year freedom rates from aortic complications, limb complications, and mortality between low-risk and high-risk groups (p<0.001).
Monsalve-Torra A, et al. [13]	2016	The study employed various ML techniques including multilayer perceptron, radial basis function, and BNs to develop a predictive system for in-hospital mortality in patients undergoing open repair of AAA.	Multilayer perceptron, radial basis function, BNs	57 attributes from 310 cases	The examined algorithms showed over 91% accuracy, but sensitivity and specificity varied. Feature selection improved performance, particularly for RBF and BN algorithms. The highest sensitivity for death prediction was 86.8%, with specificity between 96.8% and 98.6%. Combining the three algorithms notably increased the sensitivity of mortality rate prediction.
Hadjianastassi-ou, et al. [14]	2006	The study included 1205 elective and 546 emergency AAA patients, using four independent physiological variables to predict in-hospital mortality. Both multiple regression and ANN models were developed, trained on 75% of the patient population, and tested on the remaining 25%. The evaluation included calibration, discrimination, and comparison with clinicians' estimates.	ANN	1205 elective surgery patients and 546 emergency surgery patients	In-hospital mortality rates were 9.3% for elective surgery (95% CI: 7.7%-11.1%) and 46.7% for emergency surgery (95% CI: 42.5%-51.0%). Both the ANN and statistical models outperformed clinicians' predictions in accuracy. However, only the statistical model maintained internal validity in the validation set, with good calibration (Hosmer-Lemeshow C statistic: 14.97; P=0.060) and discrimination (AUC: 0.869; 95% CI: 0.824-0.913).
Kodenko, et al. [18]	2022	The review assesses opportunistic screening models, specifically the interpretation of noncontrast CT scans including the abdominal aorta. The index test, evaluating AAA detection via AI algorithms, requires fully automatic segmentation of noncontrast CT images. Manual expert segmentation serves as the reference standard, evaluated using agreement metrics or observer expertise level.	Digital image processing algorithms, Hough’s algorithm, NN, and non-NN logical algorithm	355 cases from eight studies, with 273 cases (77%) containing AAA	The review focused on automated AAA detection or segmentation in non-contrast abdominal CT scans. The mean sensitivity value was 95%, the mean specificity value was 96.6%, and the mean accuracy value was 95.2%. However, high study heterogeneity was observed, and further research with balanced noncontrast CT datasets and adherence to reporting standards is needed to validate the high sensitivity value obtained.
Adam, et al. [20]	2021	A neural network pipeline, trained on 489 CTAs, automates the measurement process. Validation used a separate set of 62 CTAs, including controls, aneurysmal aortas, and aortic dissections, scanned before and/or after endovascular or open repair.	ARVA	Training - 489 CTAs and validation - 62 CTAs, including controls, aneurysmal aortas, and aortic dissections	The range of median absolute differences compared to expert measurements varied from 1 mm to 2 mm across all annotators, with ARVA showing a median absolute difference of 1.2 mm.
Berman, et al. [24]	2011	An interactive decision support tool, incorporating the latest outcomes data and input from surgeons and patients, was developed and piloted with AAA repair candidates. Recruited from a university vascular surgery clinic and a VA hospital, patients used the tool before meeting their surgeons. Feasibility and acceptability were gauged by participation rates, time required, assistance needed, and patient opinions. Effectiveness was evaluated by assessing changes in knowledge and decisional conflict using paired t-tests.	Decision support tool	12 patients	All approached patients (n=12) agreed to participate in the study. The tool was used for a median duration of 35 minutes (range: 25-45 minutes), and patients navigated the program with minimal technical assistance. Knowledge scores showed a significant increase from 56% to 90% (P<0.005), while decisional conflict scores decreased from 29% to 8% (P<0.04). Patients reported that the tool provided balanced information across treatment options, presented information clearly, helped them organize their thoughts, and prepared them for discussions with their surgeon.