Project 462960

The number of women diagnosed with Ductal Carcinoma In Situ (DCIS), a non-invasive form of breast cancer, has increased rapidly since the introduction of breast screening programs. DCIS is regarded as a potential precursor to invasive cancer so treatment typically involves surgery to remove the lesion. However, only 14-53% of DCIS cases progress to invasive cancer over a period of 10 or more years and there are growing concerns about overtreatment. This has led to attempts to de-escalate treatment by carrying out less invasive surgery, or even omitting surgery altogether. Several trials are ongoing to investigate whether it is safe to manage women with low-risk DCIS with active surveillance. To safely de-escalate treatment, it is essential to rule out the presence of invasive disease. Unfortunately, it has been found that in 18-35% of cases assessed as DCIS on the initial biopsy, invasive cancer is found in the surgically removed tissue - this is known as upstaging. When upstaging occurs, the patient may have to be recalled for a second procedure to carry out a lymph node biopsy since the risk of metastases can no longer be ruled out. This is costly, has an increased risk of complications and causes stress to the patient. For surveillance trials, the risk of upstaging is a very significant problem since surgery is delayed until invasive cancer is detected, increasing the risk of progression. An accurate method of predicting upstaging is essential to ensure that women with occult invasive disease receive appropriate treatment whilst still ensuring that women with purely non-invasive DCIS are not over-treated. We will train an artificial intelligence algorithm to predict upstaging using information extracted from mammograms and images of biopsy specimens. This will provide a personalized estimate of the probability that an initial diagnosis of DCIS will be upstaged to invasive cancer, allowing each woman to make a more informed choice about treatment options.