In a study of more than 100,000 screening mammograms, researchers showed the poteial of an artificial ielligence tool to ideify women at high risk of dista breast cancer. An ierval cancer is one that is diagnosed between two regular screening mammograms. The results of this new study in Published radiology journal is
According to the medical technology section of Tecna news media, dista cancers generally have a worse prognosis compared to cancers that are detected during screening. These cancers are usually more aggressive or larger. For this reason, Fiona J. Gilbert, one of the authors of this study, emphasized the importance of minimizing the number of these cancers in any screening program.
Gilbert and lead researcher Joshua WD Rothwell used large retrospective data from the UK triennial screening programme. They used artificial ielligence to ideify women who needed additional imaging to find dista cancers. Personalized screening depends on an accurate assessme of cancer risk over a specific time period, Gilbert explained.
This study included 134,217 screening mammograms from the same number of women aged 50 to 70, among which there were 524 ierval cancers. Negative digital mammograms were processed by a deep learning algorithm called Mirai. The AI tool created an overall risk score for ierval breast cancer, mainly using information from the mammogram itself, such as tumor characteristics and breast density.
The three-year risk scores of this AI tool retrospectively predicted a significa perceage of dista cancers. For women who were in high-risk groups of 1%, 5%, 10%, and 20%, this tool predicted 3.6%, 14.5%, 26.1%, and 42.4% of all 524 dista cancers, respectively.
Rothwell noted that the results show that more screening of mammograms in the top 20% of scoring could detect 42.4% of dista cancers. This means that Mirai can be used to ideify women for additional imaging or to shorten the screening ierval. This method can replace or suppleme breast density assessme.
The AI tool performed better in predicting dista cancers during the first year after screening. Although this tool was less effective in women with very dense breast tissue, its performance was still superior to conveional risk prediction tools. Artificial ielligence can help optimize screening programs in couries like the UK.
This optimization could be through improving the selection criteria of women who benefit from complemeary imaging such as MRI or mammography with corast. Of course, Gilbert poied out that calling 20% of women for complemeary imaging will require significa capacity building to provide these services.
Next steps for researchers include comparing commercially available predictive AI tools, performing economic modeling and cost-effectiveness analysis. They also plan to conduct a clinical trial to ideify women who would benefit most from additional imaging. The ultimate goal is to accurately ideify high-risk women while minimizing the volume of additional imaging.
