Warren Krackov1, Murat Sor2, Rishi Razdan2, Hanjie Zheng3, Peter Kotanko3,4. 1. Azura Vascular Care, Malvern, Pennsylvania, USA, Warren.Krackov@AzuraCare.com. 2. Azura Vascular Care, Malvern, Pennsylvania, USA. 3. Renal Research Institute, New York, New York, USA. 4. Icahn School of Medicine and Mount Sinai, New York, New York, USA.
Abstract
BACKGROUND: Innovations in artificial intelligence (AI) have proven to be effective contributors to high-quality health care. We examined the beneficial role AI can play in noninvasively grading vascular access aneurysms to reduce high-morbidity events, such as rupture, in ESRD patients on hemodialysis. METHODS: Our AI instrument noninvasively examines and grades aneurysms in both arteriovenous fistulas and arteriovenous grafts. Aneurysm stages were adjudicated by 3 vascular specialists, based on a grading system that focuses on actions that need to be taken. Our automatic classification of aneurysms builds on 2 components: (a) the use of smartphone technology to capture aneurysm appearance and (b) the analysis of these images using a cloud-based convolutional neural network (CNN). RESULTS: There was a high degree of correlation between our noninvasive AI instrument and the results of the adjudication by the vascular experts. Our results indicate that CNN can automatically classify aneurysms. We achieved a >90% classification accuracy in the validation images. CONCLUSION: This is the first quality improvement project to show that an AI instrument can reliably grade vascular access aneurysms in a noninvasive way, allowing rapid assessments to be made on patients who would otherwise be at risk for highly morbid events. Moreover, these AI-assisted assessments can be made without having to schedule separate appointments and potentially even via telehealth.
BACKGROUND: Innovations in artificial intelligence (AI) have proven to be effective contributors to high-quality health care. We examined the beneficial role AI can play in noninvasively grading vascular access aneurysms to reduce high-morbidity events, such as rupture, in ESRD patients on hemodialysis. METHODS: Our AI instrument noninvasively examines and grades aneurysms in both arteriovenous fistulas and arteriovenous grafts. Aneurysm stages were adjudicated by 3 vascular specialists, based on a grading system that focuses on actions that need to be taken. Our automatic classification of aneurysms builds on 2 components: (a) the use of smartphone technology to capture aneurysm appearance and (b) the analysis of these images using a cloud-based convolutional neural network (CNN). RESULTS: There was a high degree of correlation between our noninvasive AI instrument and the results of the adjudication by the vascular experts. Our results indicate that CNN can automatically classify aneurysms. We achieved a >90% classification accuracy in the validation images. CONCLUSION: This is the first quality improvement project to show that an AI instrument can reliably grade vascular access aneurysms in a noninvasive way, allowing rapid assessments to be made on patients who would otherwise be at risk for highly morbid events. Moreover, these AI-assisted assessments can be made without having to schedule separate appointments and potentially even via telehealth.