AI shows it can improve predictions for aggressive breast cancer

Combining AI-driven image analysis with image-based risk scoring has demonstrated the ability to predict long-term breast cancer risk, according to a new study from the Mayo Clinic and the University of California, San Francisco.

why it matters

in a report published in Journal of Clinical OncologyThe researchers used ScreenPoint Medical’s TransPara EXAM scores, along with Valpara’s TruDensity software, to study how this combination could assist radiologists in predicting the long-term risk of advanced and endstage breast cancer.

Mayo Clinic and UCSF pulled two-dimensional full-field digital mammograms taken 2–5.5 years before cancer diagnosis from 2,412 women with invasive breast cancer and 4,995 controls from two US mammography cohorts.

They found that using AI-driven image analysis with an image-based risk tool that classifies exams would enable radiologists to accurately quantify breast tissue, as image software can assess breast density. Reduces reader variability.

TruDensity AI algorithms generate accurate volumetric measurements of breast structure using a combination of X-ray physics and machine learning to eliminate variability arising from human interpretation, according to an announcement about the new US study.

“Breast density is an important factor in assessing breast cancer risk, and an objective, volumetric measurement of density is important,” Dr. Ralph Highnam, chief science and innovation officer at Volpara Health, said in a statement.

“Through the power of AI, we can uncover valuable insights that help clinicians identify individuals at risk for cancer and design personalized screening and prevention strategies.”

This retrospective study echoed a previous study published in March European Radiology that merged the screening results of women participating in BreastScreen Norway with AI scores and determined consensus for different theoretical scenarios of AI and radiologists in screen-reading, recall, and cancer detection.

The Norwegian study tested the approach using TransPara’s test scores and TruDensity’s image analysis to evaluate 949 screen-detected breast cancers, 305 interval cancers and 13,646 negative exams between 2010 and 2018.

“We assessed breast imaging reporting and data system density, AI malignancy score (1-10), and volumetric density measures,” the US researchers said in their abstract.

“We used conditional logistic regression to estimate the odds ratio, 95% CI, adjusted for age and BMI and the C-statistic, to estimate the AI ​​score with invasive cancer and its contribution to the model with breast density measures,” he added. can be described.”

While the AI ​​score improved prediction of all cancer types in models with density measures, and improved discrimination for advanced cancer, the approach did not reach statistical significance for interval cancer, or breast cancer detected among mammography screenings. walked.

Norwegian researchers note that 20–30% of screen-detected and interval cancers are classified as misses in retrospective informed review studies, and cite double reading protocols and a lack of European radiologists to use AI-powered diagnostic Increased tool requirement.

“AI systems have been proposed as a tool to support or replace radiologists in the reading process,” they wrote in the full report. They were able to compare the AI’s performance against the radiologist’s double readings.

“The accuracy of the AI ​​system was comparable to that of a reader in an independent double-reading setting,” he said.

big trend

For many years, the fight against breast cancer has been complicated by mammogram readings in women with high levels of breast tissue density. The campaign aims to advance early detection and get women to care faster – and to quell rising breast cancer mortality rates around the world.

Breast cancer isn’t the only area where artificial intelligence could help radiologists, and many fear AI will replace doctors.

While the immediate and long-term benefits of AI can advance a diagnosis and then drive care-management decisions, providers should rest assured, Dr. According to Michael Donovan, cofounder and chief medical officer at PreciseDx, a health IT company that uses AI to personalize medicine.

“An important point is that, through these benefits and competencies of care that AI can provide to the physician, the final arbiter, from a clinical, ethical and medico-legal perspective, is the physician,” he pointed out. Healthcare IT News In January.

“For example, AI cannot make a final diagnosis on a radiographic scan alone or provide a definitive diagnosis on an image of a needle-biopsy sample of a patient.”

On the record

“While we’ve known for decades that density and breast cancer risk are correlated, recent research has really advanced our ability to better understand the effects of density when combined with image-based risk to drive personalized medicine for women.” further,” Dr. Nico Karsmeijer, ScreenPoint Medical’s chief scientific officer and professor at the University of Radboud, said in the statement.

Andrea Fox is a senior editor for Healthcare IT News.
Email: afox@himss.org
Healthcare IT News is a HIMSS Media publication.