Breast cancer: with AI, mammograms can bring forward diagnosis by up to six years

It’s a well-known fact. When it comes to cancer, early diagnosis is crucial for a better prognosis. This is particularly true for breast cancer, to the extent that screening is a fundamental public health strategy for women’s health. In short, mammography is essential. And perhaps, thanks to Artificial Intelligence, it could become even more decisive, identifying cancerous lesions well before a specialist radiologist’s eye could detect them. In fact, according to Swedish research published in *Radiology*, three different AI systems specifically developed for this examination are reportedly capable of detecting signs of cancer up to six years before diagnosis. In particular, the three AI-based computer-aided detection systems (AI-CAD), using mammography data from a large screened population, yielded, on average, higher tumour prediction scores for those who were subsequently diagnosed with breast cancer, compared with lower scores for those who did not develop the disease.

The study examined 88,963 mammograms carried out on 31,394 patients over a 10-year period, using data extracted from the VAI-B (Validation of Artificial Intelligence for Breast Imaging) database, which collects information on examinations across four regions of Sweden. The Swedish national mammography screening programme invites women aged between 40 and 74 to attend screening examinations every two years, and each mammogram is traditionally interpreted by two radiologists.

The study, coordinated by Fredrik Strand of the Karolinska University Hospital in Stockholm, applied three commercially available artificial intelligence-based CAD (AI-CAD) systems to mammograms taken between January 2008 and April 2019. During that period, 12,072 participants – 38.5 per cent – were diagnosed with cancer by radiologists. The AI-CAD systems successfully identified many of these cancers at early screening stages, achieving a specificity of 90 per cent – distinguishing between a true positive and a true negative result – in up to 19.7 per cent of women 6 years prior to their recorded diagnosis, up to 25.2 per cent 4 years prior to diagnosis, and up to 39.3 per cent 2 years prior to diagnosis.

In future, therefore, AI-CAD scores could help radiologists to detect, at an early stage, mammographic signs of potential future tumours, and a personalised approach to interpreting individual scores could help to identify patients who might benefit from closer monitoring. “Around 20 per cent of breast cancer cases show mammographic signs that are already visible to AI some six years before diagnosis,” Strand commented in a statement. “Our study confirms the potential of AI to detect, in some cases, signs of cancer on mammograms long before they are detected by radiologists.”

Undoubtedly, experts believe that artificial intelligence offers extraordinary potential in breast diagnostics. “It is, however, important to clarify one point: identifying retrospectively, using AI, mammographic signs associated with the future development of a carcinoma does not necessarily mean being able to diagnose that tumour six years earlier,” notes Massimo Calabrese, Director of the Oncology and Breast Radiology Unit at the IRCCS Azienda Ospedaliera Metropolitana (AOM) – Policlinico San Martino in Genoa – points out. “The study is certainly interesting and opens up important prospects, but the results must be interpreted with caution, also bearing in mind the retrospective nature of the research. Above all, it will be necessary to demonstrate that this predictive capability can translate into a real clinical benefit for women”.