AI-Enhanced Mammograms Lower Risk of Malignant Breast Cancer

Recent studies indicate that women screened for breast cancer with AI-assisted radiology experience a significant reduction in the development of advanced cancer by their next screening compared to those assessed by a traditional radiologist alone, sparking hopes that AI technology could enhance patient outcomes.

“This is the first randomized controlled trial examining AI’s effectiveness in mammography screening,” states Christina Lång from Lund University, Sweden.

The AI-assisted method utilizes advanced software trained on over 200,000 mammography scans from 10 countries to evaluate the likelihood of cancer on a 1 to 10 scale based on distinctive visual patterns in the scans. Scans rated 1 to 9 are reviewed by a single experienced radiologist, while those with a score of 10, indicating a high likelihood of cancer, are assessed by two radiologists for a more thorough evaluation.

Previous research has shown that the AI approach can identify 29% more cancers compared to standard evaluations, where two radiologists review each mammogram without increasing the false-positive rates. “That’s truly impressive,” notes Fiona Gilbert, a doctor at Cambridge University who was not involved in the study.

Furthermore, Lång and her team have discovered that the AI approach significantly lowers the incidence of interval cancers—tumors that develop rapidly between regular screenings, making them particularly aggressive and prone to metastasis.

The study involved over 100,000 Swedish women aged 55 and older, with roughly half receiving standard breast cancer screening reviewed by two radiologists, while the other half were screened using an AI model developed by ScreenPoint Medical, with results evaluated by an experienced radiologist in Nijmegen, Netherlands.

Women who benefited from AI-assisted screening were, on average, 12% less likely to develop interval cancers compared to their counterparts undergoing standard screening. “We were thrilled when the results arrived,” Lang stated.

This improved outcome could be attributed to AI’s superior ability to detect cancer at its nascent stage compared to traditional methods, ensuring that even minor tumors that could escalate into interval cancers are identified promptly.

However, Lång emphasizes that this study primarily aimed to assess whether AI performs comparably to standard screenings, not necessarily to determine if it is superior, indicating that additional research is essential to validate AI’s efficacy.

The research did not assess performance across various ethnic groups, an area that current clinical trials in the UK aim to explore, according to Gilbert.

Moreover, further studies should investigate whether less experienced radiologists achieve similar benefits using AI-assisted technology, although Gilbert does not anticipate significant differences.

Following these promising results, there are plans to implement the AI approach in southwestern Sweden within a few months, while similar trials across other nations may take up to five years to assess the approach’s adaptability to diverse populations and screening frequencies, Gilbert noted.

Establishing the cost-effectiveness of the AI model is also critical. Current estimates suggest that if AI impacts screening positively, it may justify the investment, potentially reducing interval cancer incidences by at least 5%. Radiologists will require training; however, Lång believes that the simplicity of the software will facilitate this process.

It is vital to understand that even with advancements in AI technology, radiologist involvement remains essential in breast screenings. “Women participating in screenings prefer a human touch alongside AI, and I concur; it is crucial for radiologists to utilize AI as a supportive tool,” Lång emphasizes.