O-050 - APPLICATION OF DEEP LEARNING FOR THE CREATION OF A PREDICTIVE MODEL TO STRATIFY PATIENTS UNDERGOING EVAR.

TOPIC:
Abdominal Aortic Aneurysms
AUTHORS:
La Corte F. (Unit of Vascular Surgery, Department of Biomedical and Dental Sciences and Morphofunctional Imaging, Policlinico G. Martino, University of Messina. ~ Messina ~ Italy) , Spinelli D. (Unit of Vascular Surgery, Department of Biomedical and Dental Sciences and Morphofunctional Imaging, Policlinico G. Martino, University of Messina. ~ Messina ~ Italy) , Barberi E. (Department of Engineering, University of Messina. ~ Messina ~ Italy) , Cucinotta F. (Department of Engineering, University of Messina. ~ Messina ~ Italy) , Toro D. (Unit of Vascular Surgery, Department of Biomedical and Dental Sciences and Morphofunctional Imaging, Policlinico G. Martino, University of Messina. ~ Messina ~ Italy) , Benedetto F. (Unit of Vascular Surgery, Department of Biomedical and Dental Sciences and Morphofunctional Imaging, Policlinico G. Martino, University of Messina. ~ Messina ~ Italy)
Introduction:
Follow-up protocols after EVAR differ between Centers. Anyway, the latest ESVS guidelines highlight the importance of the first visit in order to stratify patients according to the risk of late complications. Deep learning methods find applications in several scientific fields. In vascular surgery, they are used mostly in other districts both for diagnostic and predictive purposes. Aim of this work is to use Deep Learning methods to create a predictive model able to stratify patients undergoing EVAR based on the elaboration of pre-operative CT angiography, in order to be able to tailor the follow-up protocol.
Methods:
We retrospectively collected data of 125 patients who underwent EVAR at our Center between 2013 and 2020. Inclusion criteria were: availability of the pre-operative CT angiography, which had a slice thickness ≤ 3 mm; standard EVAR procedure; completion of 24 months of follow-up. Follow-up was based on DUS/CEUS examination at 1, 3, 6 and 12 months and annually thereafter, while CTAs were performed annually, as well as in case of aneurysm sac enlargement or other suspected complications. What is new in this work is about the use of pre-operative imaging to obtain 3D reconstructions in form of point clouds resembling the aortic wall surface (Figure 1) which were used as input data for a multi-layer Convolutional Neural Network and the use of a sac shrinkage classifier. Four outcomes were analysed: C1, pre-operative ruptured/symptomatic setting; C2, type II endoleak during follow-up; C3, composite outcome of type I or III endoleak, correction reintervention, aortic-related mortality; C4, residual sac shrinkage. The Neural Network was tested on its ability to correctly predict the four outcomes for each point cloud. Accuracy, precision, sensitivity and specificity were computed.
Results:
125 patients were included (mean age of 75.5 years; 97.6% male, 2.4% female). 8 cases of type I endoleak, 3 cases of type III endoleak and 33 cases of type II endoleak were diagnosed during follow-up. 12 patients underwent endoleak correction re-intervention. In 26 cases the composite outcome identified by the C3 classifier was found during follow-up. 65 cases of sac shrinkage were identified during follow-up (Table 1). The Neural Network had: 70% accuracy, 75% precision, 60% sensibility and 80% specificity for C1; 80% accuracy, precision, sensitivity and specificity for C2; 70% accuracy, 66.7% precision, 80% sensibility and 60% specificity for C3; 70% accuracy, precision, sensibility and specificity for C4 (Table 2).
Conclusion:
Higher numbers are required in order to let the Neural Network study and identify other anatomical parameters that may influence the outcomes. Deep Learning methods can't be substitutive to the decisional role of the physician which should always be based on latest scientific evidence and on clinical practice. On the other hand, these methods can give new information, improve current knowledge and provide new items for the evaluation of this pathology.
References:
1. Wanhainen A, Verzini F, Van Herzeele I, Allaire E, Bown M, Cohnert T, et al. Editor's Choice - European Society for Vascular Surgery (ESVS) 2019 Clinical Practice Guidelines on the Management of Abdominal Aorto-iliac Artery Aneurysms. European Journal of Vascular and Endovascular Surgery 2019:8-93. Doi: 10.1016/j.ejvs.2018.09.020. 2. de Mik SML, Geraedts ACM, Ubbink DT, Balm R. Effect of Imaging Surveillance After Endovascular Aneurysm Repair on Reinterventions and Mortality: A Systematic Review and Meta-analysis. J Endovasc Ther 2019;26(4):531-41. 3. Mohammadi S, Mohammadi M, Dehlaghi V, Ahmadi A. Automatic Segmentation, Detection, and Diagnosis of Abdominal Aortic Aneurysm (AAA) Using Convolutional Neural Networks and Hough Circles Algorithm. Cardiovasc Eng Technol 2019;10(3):490-9. 4. Kordzadeh A, Hanif MA, Ramirez MJ, Railton N, Prionidis I, Browne T. Prediction, pattern recognition and modelling of complications post-endovascular infra renal aneurysm repair by artificial intelligence. Vascular 2021;29(2):171-82. 5. Hahn S, Perry M, Morris CS, Wshah S, Bertges DJ. Machine deep learning accurately detects endoleak after endovascular abdominal aortic aneurysm repair. JVS: Vascular Science 2020:5-12. Doi: 10.1016/j.jvssci.2019.12.003. 6. Ding N, Hao Y, Wang Z, Xuan X, Kong L, Xue H, et al. CT texture analysis predicts abdominal aortic aneurysm post-endovascular aortic aneurysm repair progression. Scientific Reports 2020. Doi: 10.1038/s41598-020-69226-1.
ATTACHMENTS: