Can Machine Learning Enhance Ultrasound Differentiation Between Malignancy and Benign Lymph Node Changes Caused by COVID-19 Vaccination?

When it comes to differentiating between malignant lymph node changes and benign lymph node changes due to COVID-19 vaccination, combining a machine learning model with ultrasound may be more beneficial than visual ultrasound assessment by radiologists.

The retrospective study, recently published in the European Journal of Radiology, examined a machine learning model with ultrasound images drawn from patients who had routine breast and axillary lymph node assessment between November 2018 and February 2020 at a hospital in Barcelona. The study authors noted that the ultrasound images, acquired by breast imaging radiologists, came from patients with breast cancer as well as patients without breast cancer.

After training of the deep learning model with 10 cases and 61 controls, the study authors compared the effectiveness of the model to visual lymph node assessment by radiologists with a review of 109 ultrasound images from 36 cases and 73 controls. Visual inspection of axillary lymph nodes via ultrasound by radiologists resulted in 69.7 percent accuracy, 41.7 percent sensitivity and 83.6 percent specificity, according to the study. In contrast, the study authors noted that use of the deep learning model in concert with ultrasound led to a 92.7 percent accuracy rate, 77.8 percent sensitivity and 100 percent specificity.

“This direct comparison on the same images clearly suggests that the proposed methods can improve radiologists’ performance and also indicate that visual techniques are not enough to correctly classify nodes in patients after COVID-19 vaccination,” maintained Xavier P. Burgos-Artizzu, BEng, MSc, PhD, who is affiliated with the Barcelona Center for Maternal-Fetal and Neonatal Medicine at the Hospital Clinic de Barcelona in Spain, and colleagues.

The study authors noted that software incorporating the deep learning algorithm could prevent unnecessary biopsies and augment decision-making for patients with suspicious lymph nodes revealed during ultrasound exams.

Use of the deep learning software may also facilitate non-invasive assessment of axillary lymph nodes after a patient has been treated with neoadjuvant chemotherapy, suggested Burgos-Artizzu and colleagues.

In regard to study limitations, the study authors acknowledged the small numbers in the data set used to assess the machine learning model and maintained that large multicenter trials are needed to validate the results of this study.