Enterprise AI Analysis: Computer Vision Applications in Vascular Surgery

For carotid artery stenosis, several attempts to quantify the degree of stenosis with CV have been made. Segmentation was the focus of 74.4% of studies where most outlined the plaque itself or lumen-intima border and media-adventitia border in order to calculate the intimal-medial thickness, plaque area, degree of stenosis, or risk of rupture afterwards. Others characterized the tissue composition of the plaque: separating hemorrhage, calcifications, necrosis, fibrous tissue or loose matrix. Some studies (18.6%) made these diagnoses directly: either classifying the tissue composition of plaque, providing a risk assessment of the stenosis, or predicting whether the patient is symptomatic. The vast majority of the studies used ultrasound (73.3%) but MRI (14.0%), CT (8.4%) and X-rays, including dental radiographs (3.5%), were chosen as well.

Foot ulcer studies have been another major focus of CV applications in VS. The primary focus of this literature was diabetic foot ulcers but other types were considered as well. The majority of studies (54.8%) were aimed at diagnosis. Variations included identifying ischemia or infection, or gangrene; differentiating ulcers such as diabetic, pressure, surgical or venous ulcers; or scoring according to the photographed wound assessment tool (PWAT). Some studies (38.4%) also focused on delineating the wound boundaries. Only four studies made wound measurements such as length, width or area. A single study attempted to use CV to synthesize wound images of venous leg ulcers using GAN in order to train future models. Skin images were most commonly used (68.5%) followed by thermography (27.4%), which makes use of the altered temperature distributions in foot soles with lower temperatures observed in ulcerated areas.

The field of computer vision in medicine has seen tremendous improvements in the past decade. CV, a subfield of AI defined as machine learning (ML) methods applied strictly to image inputs, has evolved rapidly with image classification and segmentation tasks achieving medical expert performance levels. Due to the advent of powerful computers, these models are able to train on large amounts of data and with new publicly available datasets, they have been growing rapidly. As a result, CV applications are being widely studied in medicine and are beginning to be used in clinical settings. The fields which have observed the largest impacts are those whose diagnostic tasks rely on pattern-recognition and have large amounts of images available. A prime example being radiology. However, CV applications are also being studied in other fields.