Breast cancer detection extended
Print
AI breast cancer detection could predict disease four years in advance.
Artificial intelligence has shown the potential to detect breast cancer long before conventional diagnostic methods, according to a study by researchers from Norway and the United States.
The study, which involved over 116,000 women, tested AI's ability to predict cancer development using historical mammogram data.
The researchers examined mammograms from 116,495 women aged 50 to 69 who had no prior breast cancer history.
These images, taken in 2004, were analysed using AI tools, and the women were tracked until 2018 to monitor cancer development.
The AI models provided continuous cancer detection scores, ranging from 0 to 100, where higher scores indicated a greater likelihood of cancer being present.
According to the study, the AI system flagged increased risk in breasts that eventually developed cancer as early as four to six years before a formal diagnosis was made.
“AI scores were higher for breasts developing versus not developing cancer 4 to 6 years before their eventual detection,” the researchers note.
The findings suggest that AI could play a crucial role in identifying high-risk individuals, allowing for early diagnosis and potentially improving treatment outcomes.
The study used a commercial AI tool designed for cancer detection, although it was not originally intended for predicting future cancer risks.
Despite this, the AI system demonstrated an accuracy that rivals established risk calculators like the Tyrer-Cuzick model, which evaluates a range of clinical risk factors for breast cancer.
The researchers suggest that AI could be integrated into existing screening protocols to identify high-risk patients earlier and recommend supplemental screening.
Although promising, the study also acknowledges limitations.
The AI models showed a greater increase in accuracy closer to the time of cancer detection, particularly for screening-detected cancers, while interval cancers - those that develop between regular screenings - were harder to detect early.
The researchers call for further studies to refine AI's predictive capabilities and reduce false positives.
They aim to develop models combining AI scores with clinical factors such as age and mammographic density, which could personalise cancer risk assessments for women undergoing routine screenings.
The full study is accessible here.
