Leveraging Machine Learning and Artificial Intelligence to Improve Peripheral Artery Disease Detection, Treatment, and Outcomes

Abstract

Peripheral artery disease is an atherosclerotic disorder which, when present, portends poor patient outcomes. Low diagnosis rates perpetuate poor management, leading to limb loss and excess rates of cardiovascular morbidity and death. Machine learning algorithms and artificially intelligent systems have shown great promise in application to many areas in health care, such as accurately detecting disease, predicting patient outcomes, and automating image interpretation. Although the application of these technologies to peripheral artery disease are in their infancy, their promises are tremendous. In this review, we provide an introduction to important concepts in the fields of machine learning and artificial intelligence, detail the current state of how these technologies have been applied to peripheral artery disease, and discuss potential areas for future care enhancement with advanced analytics.

Keywords: artificial intelligence; deep learning; machine learning; peripheral artery disease; precision medicine; vascular disease.

Figure 1:. Applications of AI and ML across the spectrum of peripheral artery disease care, spanning diagnosis, risk stratification, management, and intraoperative guidance.

AI, artificial intelligence. EHR, electronic health record. ML, machine learning. PAD, peripheral artery disease. (Illustration credit: Ben Smith).

Figure 2:. Framework for unsupervised cluster analysis and types of deep neural networks.

A, Unsupervised cluster analysis can reveal clinically relevant patterns in data in a process where raw high-dimensional data is mathematically summarized and clustered. This general framework may be used to derive unique disease subtypes from patient data elements in a clinical database or the electronic health record. B, Feedforward neural network, composed of an input layer, multiple hidden layers, and an output layer. C, Convolutional neural networks capture local image features which are down-sampled in pooling layers to retain the most valuable features. Many pooled feature maps then pass through a flatten layer to form a single linear vector, thereby enabling input to fully connected layers that perform image classification. A, modified from Ross et al.

Figure 3:. When applied to the EHR, machine learning models reliably identify PAD and predict future cardiovascular events.

A, ML-based natural language processing (ML-NLP) of clinical notes identifies patients with PAD more accurately than structured data based on administrative diagnosis codes. B, By incorporating structured and unstructured EHR data, ML algorithms accurately risk stratify PAD patients at high risk for major adverse cardiovascular events (MACE). AUC, area under the receiver operating curve. EHR, electronic health record. MACE, major adverse cardiovascular event. PAD, peripheral artery disease. X-axis represents 1-specificity. Y-axis represents sensitivity. A, modified from Weissler et al. B, modified from Ross et al.

Figure 4:. Imaging assessment of carotid atherosclerosis and AAA disease using deep learning.

A, AI software enables automated analysis and reporting of carotid ultrasounds. B,C, Deep learning pipelines have also been developed for fully automated volumetric analysis of AAA disease from CT images. B, Representative axial CTA slices are shown for automated segmentation of the aortic lumen (red) and thrombus (green) and the manually extracted ground truth. Key for comparison image: true positive lumen and thrombus (green and yellow, respectively), false negatives (red), false positives (blue). C, 3D U-Net output for the inner wall and outer wall/ILT (intraluminal thrombus) along with its respective ground truth. Points of discrepancy are boxed. Percentages in B-C represent the Dice coefficient. AAA, abdominal aortic aneurysm. AI, artificial intelligence. CTA, computed topography angiography. DL, deep learning. A, modified from See-Mode AVA user interface. B, modified from Caradu et al. C, modified from Chandrashekar et al.

Figure 5:. Computer vision for PAD imaging interpretation.

A, AI systems may automate interpretation of CTAs and generate reports based on TASC II classification. B,C, Computer vision algorithms may also enable more efficient processing of MRI images without compromising accuracy. Computer vision models generate accurate calf muscle perfusion maps in less than 1 second, compared to 180 minutes by standard modeling (B, modified from Zhang et al.). Using standardized knee MRIs, fully automated pipelines for assessing atherosclerosis burden in the popliteal arteries have also been reported (C, modified from Chen et al.). AI, artificial intelligence. AK, above knee. AT, anterior tibial. BK, below knee. CFA, common femoral artery. CTA, computed topography angiography. DCE-MRI, dynamic contrast-enhanced magnetic resonance imaging. R, correlation coefficient. PT, posterior tibial. SFA, superficial femoral artery. TASC II, Trans-Atlantic Intersociety Consensus II.