AI Spots Potentially Fatal Blood Clots from Ultrasounds
Using AI to diagnose DVT clots from scans.
Posted September 23, 2021 | Reviewed by Kaja Perina
Artificial intelligence (AI) machine learning (ML) is being applied as a diagnostic tool for medical and health care uses. A new study published in npj Digital Medicine demonstrates how AI deep learning was able to distinguish healthy patients from those who have deep vein thrombosis (DVT) as accurately as the existing gold standard method from ultrasound scans.
“This study provides a proof of concept that ML-based analysis can distinguish patients with and without DVT while providing image acquisition guidance for non-experts according to the clinical standard,” wrote the study lead author Dr. Nicola Curry at Oxford University's Radcliffe Department of Medicine and Oxford University Hospitals NHS Foundation Trust, along with a team of researchers affiliated with the University of Oxford, Imperial College, the University of Sheffield, and ThinkSono.
According to the scientists, this is the first study to demonstrate the potential benefit of AI machine learning for diagnosing deep vein thrombosis based on ultrasound images.
Deep vein thrombosis is a clot in a deep vein, typically in the leg, that may cause a fatal pulmonary embolism (PE). Pulmonary embolisms are preventable and occur when a DVT clot breaks from the vein wall and then either partially or completely blocks the blood supply.
An estimated 60,000-100,000 Americans die each year from DVT/PE, also called venous thromboembolism (VTE), according to the Centers for Disease Control and Prevention (CDC). In England, hospital-acquired venous thromboembolism accounts for roughly 25,000-32,000 deaths annually according to figures from the House of Commons Health Committee report.
“Compression ultrasound of the legs is the diagnostic gold standard, leading to a definitive diagnosis,” the researchers wrote. “However, many patients with possible symptoms are not found to have a DVT, resulting in long referral waiting times for patients and a large clinical burden for specialists. Thus, diagnosis at the point of care by non-specialists is desired.”
In developed countries with, to diagnose DVT, clinicians often conduct a blood test to search for a small protein fragment called a D-dimer that is present in blood after fibrinolysis occurs, which is the enzymatic breakdown of fibrin in blood clots. Clinicians will combine the results of the blood test with a Wells score based on a questionnaire about the patient’s medical history and findings of a medical examination. To confirm a suspected DVT, the next step is to conduct an ultrasound scan. However, these types of specialized resources are often not accessible in lower-income areas and less-developed nations.
“Currently, no reliable test is available that can be used in a general healthcare setting (GP practice, community hospital, on a hospital ward) or be used remotely at the point of care (nursing home, patient’s home),” wrote the scientists. “Between 85 and 90% of patients presenting to their GP in high-income countries with a suspected DVT will be investigated only to find no evidence of a thrombus.”
To solve this challenge, the researchers developed an AI deep learning called AutoDVT, which is a convolutional neural network (CNN). They trained the AutoDVT using ultrasound videos from 255 study participants, then conducted assessments with 53 prospectively enrolled patients from the Oxford Haemophilia and Thrombosis Centre adult DVT clinic, and 30 prospectively enrolled patients from the Ernst von Bergmann Klinikum Potsdam clinic in Germany.
“We collect images in a pre-clinical study and investigate a deep learning approach for the automatic interpretation of compression ultrasound images,” the researchers concluded. “Our method provides guidance for free-hand ultrasound and aids non-specialists in detecting DVT.”
This machine-learning diagnostic tool in combination with cost-effective mobile ultrasound devices has the potential to enable front-line health care workers and non-specialists to perform DVT screening. This is helpful in rural clinics and understaffed hospitals that lack DVT specialists. Additionally, this technology enables specialists such as radiologists and sonographers to work remotely, rather than administer the scans themselves.
According to the scientists, theirs is a pioneering study that is proof-of-concept that AI machine learning can diagnose deep vein thrombosis from ultrasound scans—a discovery with the potential to transform the future of health care with more accessible, cost-effective diagnostics.
Copyright © 2021 Cami Rosso All rights reserved.