LAUSR

INTRODUCTION Lower extremity arterial Doppler (LEAD) and duplex carotid ultrasound studies are used for the initial evaluation of peripheral arterial disease (PAD) and carotid stenosis. Intra and inter lab variability may exist between interpreters and other interpreter responsibilities may delay the timeliness of the report. In order to address the current deficits, we examined whether machine learning algorithms could be used to classify these Doppler studies. METHODS We developed a hierarchical deep learning model to classify aortoiliac disease, femoropopliteal and trifurcation disease in LEADs and developed a random forest machine learning algorithm to classify the amount of carotid stenosis from duplex carotid ultrasound using experienced physician interpretation in an active, credentialled vascular lab as a gold standard. Waveforms, pressures, flow velocities and the presence of plaque were input into a hierarchal neural network. AI was developed to automate the interpretation of these LEAD and carotid duplex ultrasound studies. Statistical analysis was done using the confusion matrix. RESULTS 5,761 LEAD studies from 2015 to 2017 and 18,650 duplex carotid ultrasound studies from 2016 to 2018 were extracted from the IU Health system. The results show the ability of artificial intelligence algorithms and methodology with a 97.0% accuracy for predicting normal cases, an 88.2% accuracy for aortoiliac disease, a 90.1% accuracy for femoropopliteal disease and a 90.5% accuracy for trifurcation disease. For ICA stenosis the accuracy was 99.2% for predicting 0-49% stenosis, 100% for predicting 50-69% stenosis,100% for predicting greater than 70% stenosis, and 100% for predicting occlusion. For CCA stenosis, the accuracy was 99.9% for predicting 0-49% stenosis, 100% for predicting 50-99% stenosis, and 100% for predicting occlusion. CONCLUSIONS The machine learning models using LEAD data, based on the collected blood pressures, waveforms; and duplex carotid ultrasound data based on flow velocities and the presence of plaque shows that novel machine learning models are reliable in differentiating normal from diseased arterial systems and accurate in classifying the extent of vascular disease.