Histological Segmentation of the Superficial Femoral Artery From Microscan to CT Using Artificial Intelligence

Overview

The femoropopliteal artery segment (FPAS) is one of the longest arteries in the human body, undergoing torsion, compression, flexion and extension due to lower limb movements. Endovascular surgery is considered to be the treatment of choice for the peripheral arterial disease, the results of which depend on the physiological forces on the arterial wall, the anatomy of the vessels and the characteristics of the lesions being treated. The atheromatous disease includes, in a simple way, 3 categories of plaques: calcified, fibrous, and lipidic. The study of these plaques and their differentiation in imaging and histology in the FPAS has already been the subject of research. To treat them, there are angioplasty balloons and stents with different designs and components, with different mechanical properties and different impregnated molecules. There is no non-invasive method (imaging) to accurately differentiate lesions along the FPAS. The analysis is performed from the preoperative CT scan, but there are high-resolution scanners that allow a quasi-histological analysis of the tissue. This microscanner can be used ex vivo. In the framework of a project, the learning algorithm was be créated (Convolutional Neural Networks) to automatically segment microscanner slices: after taking FPAS from amputated limbs, we correlated ex-vivo microscanner images of the arteries with their histology. The correlation was then performed manually between the microscanner images, and the histological sections obtained. the algorithm well be trained on these slices and validated its performance. The validation of the CT and microscanner concordance was the subject of scientific publications.

The aim of this study is to evaluate the technical feasibility of histological segmentation by the FPAS algorithm from CT. The results of this study will provide initial data to evaluate the interest of a subsequent larger scale study to validate the diagnostic capabilities of automated segmentation