A recent breakthrough development in medical imaging has been the emergence of point-of-care ultrasound (POCUS) devices, especially those systems that are wireless. Seen at right is a POCUS system from Clarius used to study the breast where the images appear, in real time, on a tablet screen. One of the challenges POCUS users face is to control multiple imaging functions such as gain and depth, and so Clarius has added an artificial intelligence (AI) feature based on voice recognition that frees up the user’s hands while performing an examination.

Recognising that most low- and middle-income countries lack access to breast screening programmes, meaning women with palpable lumps may wait months for diagnostic assessment, Wendie Berg and her colleagues sought to address the problem. They set out “to demonstrate that artificial intelligence (AI) software applied to breast ultrasound images obtained with low-cost portable equipment and by minimally trained observers could accurately classify palpable breast masses for triage in a low-resource setting.” Published in Radiology this week, this was a prospective study conducted over three-and-a-half years in Mexico where the participants had at least one palpable mass.

Orthogonal images were obtained first with the Vscan POCUS system from GE HealthCare, and then with a larger, standard-of-care (SOC) device manufactured by Hitachi. There were 758 masses found in 300 women and all images were analysed using the AI software from Koios Medical. The outputs were judged to be benign, probably benign, suspicious, or malignant. The mean patient age was 50 years, the average largest lesion diameter was 13mm, and 56 (7.4%) of the masses were malignant.

The figure below right (© RSNA) shows a low-grade ductal carcinoma in situ (DCIS) where A was acquired with POCUS and B with the SOC device. The AI algorithm classified the lesion as “probably benign” and “suspicious” respectively, with the difference being explained by the better quality of the SOC image. Overall, AI correctly identified 96% and 98% of women with cancer based on POCUS and SOC respectively, while SOC performed significantly better than POCUS for correctly identifying benign lesions.

Importantly, AI performance was reduced when images were obtained with the low-cost POCUS device by a minimally trained observer. Berg was nevertheless upbeat about the outcome, declaring “this should improve access, health equity, and outcomes for women.” In an accompanying editorial, Priscilla Slanetz of Boston commented: “Integrating AI into these lower-resource environments is yet another step that will lessen the burden of disease in these vulnerable patients who currently face major obstacles to care.” And that’s good news for women in the developing world.