Use of Artificial Intelligence to Predict Complications Following Abdominal Aortic Surgery

Implantation of an Endovascular Prosthesis for the Treatment of Abdominal Aortic Aneurysms: Use of a Machine Learning Model From CT Images to Predict Complications

Abdominal aortic aneurysm, a condition characterised by an increase in the diameter of the aorta, can be treated either surgically or endovascularly. In the latter, an endoprosthesis consisting of a metal spring covered with an impermeable fabric is inserted through an artery and deployed inside the aorta. This new method appears to be less invasive than surgery, but its long-term results are not yet fully understood. As a result, patients who have undergone this treatment are monitored by their surgeon to ensure that there is no endoleak. Several research teams have proposed analysing medical images to predict this risk of endoleak. Doctors are now trying to use artificial intelligence to automate the analysis of these images.

The use of endoprostheses has revolutionized the therapeutic management of patients with thoracic and/or abdominal aortic aneurysms.

In France, the placement of abdominal aortic endoprosthesis (AAE) for treating infrarenal abdominal aortic aneurysm (AAA) follows the recommendations of the French National Authority for Health (HAS) from 2001, updated in 2009 in light of medical device developments and subsequent literature data (2000-2006). It specifically states that: "endovascular treatment is less invasive than surgical treatment, allowing for reduced 30-day morbidity rates for patients eligible for both surgery and endovascular treatment." Endovascular treatment can thus be offered to patients with normal surgical risk and favorable anatomical criteria, alongside surgical treatment, after informing the patient about the benefits and risks of both methods. The vast majority of patients with AAE require regular long-term follow-up, focusing on sac dimension and volume analysis and complication management. All patients must simultaneously receive optimal treatment for their vascular risk factors and comorbidities.

The updated recommendations can be summarized as follows:

• Indications for treating asymptomatic infrarenal AAAs: AAAs with greatest diameter > 50 mm or diameter increase of more than 5 mm in six months and 10 mm in 1 year.
• A symptomatic or complicated AAA is treated regardless of size.
• Despite the absence of conclusive data, a saccular AAA is considered high-risk due to its particular morphology and is proposed for treatment even if its greatest diameter is < 50 mm.

Since 2009, the number of endoprosthetic treatments has steadily increased at the expense of open procedures, although a medical-economic evaluation has questioned the efficiency of endovascular technique in patients eligible for both techniques. The long-term benefit in preventing mortality from AAA rupture remains unestablished (the mortality benefit difference is not maintained at 4 years), and even though most patients die from their comorbidities rather than their aneurysm, the first deaths from late rupture (beyond the fifth year of implantation) have appeared in long-term follow-up studies.

Concurrently, these studies have shown that nearly half of treated patients will require one or more additional procedures in subsequent years, and among treated patients:

• Only those showing aneurysmal sac shrinkage are considered cured.
• An average volume reduction of 3.2% would be sufficient to prevent endoleak occurrence.
• Conversely, a volume increase of more than 2% would be significantly associated with endoleak existence.

Endoleak is the most common complication after AAE placement, with incidence varying greatly depending on type and time elapsed since endoprosthesis placement. It is defined by blood circulation between the AAE and the arterial wall of the AAA. Five types of endoleak have been described according to anatomical, chronological, or physiological characteristics:

• Type I Endoleak: Ia) proximal anterograde peri-prosthetic flow due to endoprosthesis fixation failure, Ib) distal retrograde peri-prosthetic flow due to endoprosthesis fixation failure, Ic) iliac occluder seal failure,
• Type II Endoleak: retrograde flow from lumbar artery(ies) or inferior mesenteric artery feeding the aneurysmal sac: IIa) reperfusion through inferior mesenteric artery, IIb) reperfusion through lumbar artery(ies)
• Type III Endoleaks: IIIa) separation of modular endoprosthesis components, IIIb) fabric degradation. Endoleaks indicating early mechanical failure or premature fatigue of the endoprosthesis, potentially facilitated by AAA morphological changes, particularly its retraction
• Type IV Endoleaks: excessive endoprosthesis porosity. Generally early (<30 days post-implantation), initially described but now almost non-existent due to fabric improvements
• Type V Endoleak: residual hyperpressure in the aneurysmal sac without obvious cause (no individualized leak) associated with AAA size increase. It warrants thorough investigation for endoleak and may lead to explantation.

Several teams have proposed preoperative criteria to predict endoleak risk (preoperative inferior mesenteric artery patency, sac size), but without real evidence. Some teams propose preventive embolizations based on these criteria (pre- or during procedure).

The investigators propose to evaluate whether a machine learning algorithm can predict endoleak risk from initial CT scan images.

Inclusion Criteria:

• Treated for an abdominal aortic aneurysm through endovascular prosthesis implantation

Exclusion Criteria:

• Underwent intraoperative embolization of the inferior mesenteric artery or aneurysmal sac
• Refuse the use of their data