Artificial intelligence may reveal individual cancer risk
“Most cancers are caused by coincidence, when ageing cells undergo a sufficient number of mutations that affect cell function,” Mannermaa says.
Besides age, many other factors, which affect the risk of breast cancer, are also known. An estimated 5–10% of cases are caused by genetic mutations that significantly increase the risk of breast cancer, such as mutations of the BRCA or PALB2 genes. There are various additional genetic factors that could raise the risk of breast cancer, alongside changes in gene regulation that can be influenced by lifestyle choices and environmental factors.
The amount of lifetime exposure to oestrogen affects at least the risk of hormone-positive breast cancer, and for example overweight and heavy drinking may increase the risk by influencing the body’s oestrogen level.
In Finland, mammography is used to regularly screen women aged 50–69 for breast cancer.
“In addition to screening for cancer, mammography images can also be used to assess the risk of breast cancer later in life. Dense breast tissue is associated with an increased risk of breast cancer.”
More detailed information on the real individual risk can be obtained by means of artificial intelligence and machine learning methods that can simultaneously examine a wide range of risk factors and their interactions. These methods are being developed by Senior Researcher Hamid Behravan, together with Mannermaa.
“For example, the genetic risk for breast cancer has previously been assessed on the assumption that different factors affect the risk independently of each other. However, we have shown that by taking their interactions into account, we can achieve an increasingly accurate risk assessment,” Behravan says.
The machine learning model developed by the group has identified genetic regions that are essential to the risk of breast cancer, and that these regions also interact with one another. The work helps to locate risk genes and to understand interactions between genes underlying the risk of cancer.
“By combining this with data on oestrogen metabolism and hereditary risk for breast cancer, we could predict cancer even more reliably.”
The results also helped to identify gene networks linked to breast cancer, which are associated with, for example, tumour growth, cell division and DNA repair mechanisms.
Behravan’s goal is to introduce a multi-factor risk assessment tool for health care professionals, which would enable the identification of both high- and low-risk individuals for breast cancer, allowing screening and monitoring to be adjusted accordingly.
“This would also help to avoid unnecessary imaging and costs.”
Artificial intelligence can also be used to interpret imaging results. The research group has developed an AI-based method for automated estimation of breast tissue density from mammography images. The method provides uniform estimates of tissue density, which may facilitate the work of radiologists, while also contributing to density assessment training.
“The reliability of this method has already been tested with thousands of mammography images obtained from Kuopio University Hospital, and the plan is to test it also in international imaging data.”
In the future, the research group strives to develop solutions not only for risk assessment but also for prognosis assessment in breast cancer patients.
“Machine learning enables the identification of patient groups that may benefit from certain treatments, for example.”