Today DM is used for breast cancer screening. In order to detect as many cancers as possible, each image is read by two specialised breast radiologists. DBT can find more cancers, but takes about twice the time to read. This is an obstacle to replace mammography with breast tomosynthesis in screening, not least because there is a shortage of breast radiologists. 

We study if a computer with artificial intelligence could help in reading the images, so it can both be faster and at the same time find more cancers. Maybe artificial intelligence can identify suspicious areas, which the radiologists should study more thoroughly in order to detect more cancers. Using AI while reading the examinations might eliminate the need for two human readers. Another alternative could be if artificial intelligence can make a preselection and make it possible to focus the radiologists’ time on the women with the highest risk of cancer. 

According to one of our studies AI can be used to identify normal cases in DM screening – leading to that about one fifth of the examinations can be excluded from human reading. Another of our studies indicated that AI might make it possible to detect additional cancers on DM – cancers that radiologists only detected on DBT. 

We have also evaluated an AI system on the DBT examinations from the MBTST and retrospectively tested different ways of utilising AI to make the reading of DBT more resource efficient. Normal cases (about 50%) could be excluded from human reading, or the second reader could be replaced by AI – both leading to a total reading time on par with double read DM screening, but detecting 25% more cancers. This could make it possible to screen all women with DBT, but keep the total reading time at the same level as when using mammography. More cancers could be detected earlier, hopefully at a curable stage, with the same resources as mammography screening.