AI-assisted ultrasounds may better detect joint bleeding, inflammation

AI models could help doctors make quicker, more accurate diagnoses, study says

Margarida Maia, PhD avatar

by Margarida Maia, PhD |

Share this article:

Share article via email
A person reclines on a couch with one leg elevated and a bag of ice on a knee.

Artificial intelligence (AI) can comb through ultrasound scans to detect joint bleeding and inflammation occurring as a result of hemophilia, a study has found, suggesting it may help doctors make quicker and more accurate diagnoses.

“AI models have the potential to aid diagnosis and enable earlier therapeutic interventions, helping [people with hemophilia] achieve healthy and active lives,” researchers wrote, noting that consistent ultrasound settings could further improve AI performance.

The study, “Artificial Intelligence-Assisted Ultrasound Imaging in Hemophilia: Research, Development, and Evaluation of Hemarthrosis and Synovitis Detection,” was published in the journal Research and Practice in Thrombosis and Haemostasis.

Hemophilia occurs when the blood cannot clot properly, leading to excessive bleeding. When bleeding occurs in the joints, it can cause inflammation and irreversible damage, making it harder for people to move and go about their day. Detecting these problems early is important for timely treatment.

Recommended Reading
Illustration of the inside of a muscle.

Study finds muscle loss common in hemophilia A with joint bleeds

AI algorithms could speed accurate diagnosis by automating detection

However, the usual method of doing so, with ultrasound scans, requires specialized skills and training. That’s where AI models or algorithms may come in. These models could help with quicker and more accurate diagnoses by automating detection.

To see if an AI algorithm could learn to detect signs of hemarthrosis (bleeding into the joints) and synovitis (inflammation of the tissue that lines the joints), researchers gathered a total of 5,649 ultrasound scans of elbows, knees, and ankles from people with hemophilia A or B in Japan.

They then used these imaging data as input to train the AI algorithm to recognize the signs of hemarthrosis and synovitis, and checked how well it could do this by comparing its results to what experts found after examining the same ultrasound scans.

Of the 5,649 ultrasound scans, 3,435 had enough quality to be used for the analysis. From these, 2,000 had normal findings, whereas 1,435 had clinical findings, according to doctors. Hemarthrosis was observed in 1,310 scans, and synovitis in 891.

After being trained to process and interpret the ultrasound scans, the AI algorithm performed relatively well in recognizing hemarthrosis, with an AUC value of 0.87 or higher, and synovitis, with an AUC of 0.90 or higher, in elbows, knees, and ankles. These values were consistent across patients from different age groups, from 10 to 60 years old.

The AUC, which stands for area under the curve, is a statistical measure that can be used to assess the diagnostic accuracy of a given test or tool, that is, its ability to correctly detect the presence of abnormal findings. AUC values range from zero to 1, with higher values indicating a higher diagnostic accuracy.

Recommended Reading
banner image for

Reflections from the Coalition for Hemophilia B symposium

Al algorithm may have identified bleeding that was originally missed

The proportion of ultrasound scans that were identified as having abnormal findings by the Al algorithm when no bleeding was recorded in the patient’s medical files was 80% or higher for hemarthrosis and 75% or higher for synovitis.

“These data suggest that the Al algorithm may have identified bleeding that was originally missed in the patient’s subjective assessment before joint imaging was performed,” the researchers wrote.

“Al models … could benefit clinical practice by reducing the burden on healthcare professionals, preventing oversights, and leading to earlier diagnoses and appropriate therapeutic intervention,” they wrote, “ultimately helping people with hemophilia to achieve a healthy and active life.”

However, “there is no clear evidence on how much control is needed for the abnormal findings detected and how to intervene,” they noted. “As such, Al models should be integrated prudently and reasonably within the practitioner’s workflow.”

This work was supported by Chugai Pharmaceutical, which developed Hemlibra (emicizumab), an approved antibody-based treatment for hemophilia A. Chugai is now part of the Roche group.