A collaboration between St. Petersburg State University of Electronic Technology and the Almazov National Medical Research Centre in Russia has produced an artificial intelligence system that automatically analyses computed tomography (CT) images and highlights regions with tumour-like features for physician review. The neural‑network software processes scans uploaded to a server, marks suspicious areas and forwards the annotated images to clinicians who make the final diagnosis.
The announcement, carried by Russian and state-linked outlets, positions the tool as an aid for early breast‑cancer detection. Early diagnosis improves outcomes in breast cancer, but the dominant screening modalities worldwide remain mammography, ultrasound and MRI; CT is rarely used as a primary breast‑screening tool because of lower soft‑tissue contrast and higher radiation dose relative to mammography.
What the Russian developers appear to be offering is an algorithm able to identify breast abnormalities on CT images that are taken for other clinical reasons or as part of chest imaging protocols. If robust, such software could convert routine CT studies into opportunistic screening opportunities — flagging lesions that might otherwise be missed and prompting targeted follow‑up with mammography or biopsy.
Crucial questions remain unanswered in the public announcement: the system’s sensitivity, specificity, false‑positive rate and performance across different scanners, patient ages and breast densities. The available report does not reference peer‑reviewed validation, prospective trials or regulatory approval, so clinicians and policy‑makers should withhold judgement until independent evaluations quantify benefits and harms.
Practical and ethical hurdles will shape adoption. Uploading patient CTs to a central server raises data‑privacy and cybersecurity concerns; clinical integration requires clear workflows so that flagged findings lead to timely, reliable follow‑up rather than unnecessary testing. There are also medico‑legal risks if automated flags are missed or if the algorithm underperforms in populations not represented in its training data.
The Russian project sits within a fast‑moving global market for AI in medical imaging, where start‑ups and hospital systems in the United States, Europe and China are already deploying or testing algorithms for cancer detection, triage and workflow optimisation. For this initiative to move beyond national publicity, its developers must publish performance metrics, run multi‑centre prospective studies and secure regulatory clearance. Only then can clinicians assess whether the tool offers a clinically meaningful advance or another unvalidated claim in a crowded field.