AI may help hemophilia patients detect joint bleeding at home
Algorithm detects bleeds using remote system
Written by |
A computer-aided diagnostic tool using artificial intelligence (AI) can accurately detect joint bleeds in adults with hemophilia, a study showed. The researchers said the findings may lead to a way for patients to obtain ultrasound images at home, rather than traveling to a facility.
“The CAD [computer-aided diagnosis] system for automatic detection of joint [bleeds] is feasible and reliable,” the team wrote. It may help with the “early recognition of joint bleeding and supporting personalized, timely therapeutic interventions to prevent further joint damage.”
The study, “Artificial Intelligence for Automated Detection of Joint Bleeding via Ultrasound in Hemophilia: Advancing Standardization,” was published in the Journal of Thrombosis and Haemostasis.
Joint bleeding is a frequent manifestation of hemophilia, a group of genetic conditions caused by deficiencies in proteins required for blood clotting. Large joints such as the knees, ankles, and elbows are most commonly affected. Repeated joint bleeds can drive chronic inflammation, which can eventually lead to hemophilic arthropathy, a painful joint disease.
Point-of-care ultrasound, or imaging performed at the patient’s location rather than in a separate facility, is a growing area of interest in hemophilia care. A remote system that allows patients to perform ultrasound imaging themselves could support faster, more accurate diagnosis of joint bleeding, enable effective treatment to begin at home, and reduce the need for clinic visits or hospitalization. It could also improve patients’ understanding of their condition and their adherence to hemophilia treatment.
Training AI to recognize patterns from ultrasound images
A challenge to developing a remote system is that it would generate a large number of images for doctors to review, making it harder to decide which cases require attention first and to make timely decisions.
Researchers at the University of Milan developed a computer-aided diagnostic tool using a deep learning algorithm, a type of AI that learns to recognize patterns from large datasets. They set out to evaluate whether it would perform accurately on a set of U.S. images.
They enrolled 158 men with hemophilia, with a mean age of 44, for their study. Nearly all (90%) were diagnosed with hemophilia A, while the rest had hemophilia B. Researchers used images collected at a hemophilia center during routine visits or for acute events such as joint bleeding. They focused on 814 ultrasound images of the subquadricipital recess (SQR), a small pouch-like space above the kneecap where fluid can accumulate.
Two specialists assessed each image using a four-level scale for capsule distension: absent, mild, moderate, or severe. Capsule distension means an abnormally stretched joint capsule, the tissue surrounding the joint, due to excess fluid, a sign of bleeding.
When the results of the two specialists were compared, the agreement was considered moderate. After the assessment was simplified to either “distended” or “non-distended,” the agreement improved substantially.
Of the 814 knee images, 340 (41.2%) showed joint capsule distension. Among those, 42 cases (12.4%) were due to bleeding within the joint, 80 (23.5%) were due to thickening of the membrane lining the joint, and 218 (64.2%) were due to an accumulation of non-blood fluid in the joint.
When researchers tested whether the AI could correctly locate the SQR within the knee, the average score across all images was 0.64, with 1.0 indicating a perfect match. The area was detected with a score above 0.5, generally considered an acceptable level of accuracy, in 80% of images.
To classify whether a joint was distended, the team defined “absent” and “mild” images as non-distended, and “moderate” and “severe” as distended. Using this approach, the model’s accuracy was 89.2% compared with the specialists’ assessment. The balanced accuracy, or the mean of true positives and true negatives, rose to 93.9%.
The team found no meaningful difference in the extent of capsule distention in the knee joint between healthy men and women matched to patients by age and sex.
The researchers said the study showed “the feasibility and good accuracy of our CAD tool,” making it “the first step of our proposed telemedicine system to achieve an early and accurate diagnosis of joint bleeding in persons with hemophilia.”
The next step, they said, will be to add elbow and ankle evaluation to the system. “When a complete CAD tool is available, we will conduct a pilot study to demonstrate the validity of images collected in a home-based setting using portable US systems,” they said.
Leave a comment
Fill in the required fields to post. Your email address will not be published.