Optimizing Breast Cancer Risk-Screening With A Dual AI Model


A team of Danish as well as Dutch researchers has gone on to combine an AI diagnostic tool along with a mammographic texture model so as to improve the evaluation when it comes to short- and long-term breast cancer risk. This novel approach goes on to represent a prominent step forward when it comes to refining the capacity to go ahead and also forecast the complexities pertaining to the breast cancer risk.

It is worth noting that one in every ten women is going to develop breast cancer at some given point in their life. The fact is that it happens to be the most common cancer in women, and goes on to get diagnosed mostly in the patients over the age of 50 years. At present, screening programs make use of mammography as their primary diagnostic tool when it comes to detecting the issue pertaining to breast cancer, and that too at an early stage, however, some lesions still go on to become quite challenging for radiologists to make sure to identify. Especially 55% of the cases happened to be accompanied by the presence of microcalcifications, which are tiny spots of calcium deposits, not over 0.1 mm in size, localized or even broadly diffused within the breast area. These calcifications happen to be frequently associated with premalignant as well as malignant lesions. In the present scenario, the majority in terms of breast cancer screening programs go on to base a woman’s estimated lifetime risk in terms of developing breast cancer on similar standard protocols.

Artificial Intelligence- AI, apparently, can be made use of for the purpose of diagnosing breast cancer earlier by way of automatically detecting the breast cancers in the mammograms and also measuring the risk which is involved in terms of future breast cancer, opined Dr. Andreas D. Lauritzen, who is a PhD, from the Department of Computer Science, University of Copenhagen. His team happened to partner along with the researchers from the Department of Radiology and Nuclear Medicine at Nijmegen’s Radboud University in the Netherlands on a project to blend two kinds of AI tools in order to leverage the respective strengths of both approaches, i.e., diagnostic models to go ahead and estimate short-term breast cancer risk as well as mammographic texture AI models to pinpoint the breast density, which happens to be an important marker when it comes to evaluating the long-term risk.

A retrospective study pertaining to the Danish women

It is well to be noted that the team of seven researchers from Denmark as well as the Netherlands looked forward to identifying whether commercially available diagnostic AI tools as well as an AI texture model, trained separately and then subsequently mixed, could go ahead and improve breast cancer risk evaluation. They made use of the diagnostic AI system Transpara, version 1.7.0, coming from a Nijmegen-based company named Screenpoint Medical B.V., and the texture model consisting of the deep learning encoder SE-ResNet 18, release 1.0, which they happened to develop themselves. Opined one of the authors of the study, Dr. My C. von Euler-Chelpin, an associate professor at the Centre for Epidemiology and Screening, Institute of Public Health, University of Copenhagen, said that a Dutch training set of more than 39,245 exams was made use of to train the deep learning models. The short- and long-term risk models happened to be combined by way of using a three-layer neural network. The mixed AI model happened to be tested on a study group of more than 119,650 women who happened to be included in a breast cancer screening program within the Capital Region of Denmark all throughout a three-year period from November 2012 to December 2015, with a minimum of five years of follow-up data. Apparently, the average age of the women happened to be 59 years.

Major interpretations show benefits

As per the results of their study, which got published in Radiology, the combination model went ahead and achieved a higher area under the curve- AUC as compared to the diagnostic AI or texture risk models separately in the case of the cancers diagnosed in two years of screening—interval cancers and those diagnosed post-this period—long-term cancers combined together.

The blended AI model also goes on to make it possible to go ahead and also identify women at high risk when it comes to breast cancer, with women identified as having the 10% highest combined risk, thereby comprising 44.1% of interval cancers as well as 33.7% of long-term cancers. According to Dr. Lauritzen and his colleagues, the findings indicated that mammography-based breast cancer risk evaluation is enhanced when blending an AI system for lesion detection as well as a mammographic texture model. Making use of AI to identify a woman’s breast cancer risk due to a single mammogram will not just result in earlier cancer detection but, at the same time, will also help alleviate strain on the healthcare system because of the worldwide shortage when it comes to specialist breast radiologists.

A fast and single mammogram approach having no clinic overheads

The present state-of-the-art clinical risk models go on to require multiple tests like blood work, mammograms, and genetic testing, as well as filling out extensive questionnaires, all of which would go on to substantially raise the workload within the screening clinic. Using their model, risk can go on to be evaluated by way of the same performance as the clinical risk models, however, within seconds of screening and even without introducing the overhead in the clinic, said Dr. Lauritzen in one of the press releases.

The fact is that the Danish-Dutch research team is now going to focus in terms of investigating the combination model architecture as well as further determining if the model adapts sufficiently to certain other mammographic devices along with institutions. The team went on to conclude that there has to be additional research which, apparently has to go ahead and focus in terms of translating combined risk to lifetime risk or even absolute risk for comparison when it comes to the traditional models.