The use of artificial intelligence (AI) in the medical field is rapidly advancing, and it is poised to mark a new milestone in the fight against breast cancer. Significant progress has been made, with scientific publications and international trials multiplying the prospects for integrating these technologies into medical practices. According to Isabelle Thomassin-Naggara, a radiology professor at Sorbonne Université and a radiologist at the Tenon Hospital, certain AI applications in breast imaging are mature enough for deployment in several European countries.
In France, during Pink October, the government has expressed interest in using AI to streamline mammogram sorting within the organized screening program. This evolution is also being considered by the French Society of Women's Imaging and the National Professional Council of Radiology, who have addressed this issue in their white paper.
A key advantage of AI lies in its ability to identify mammograms without suspicious abnormalities, demonstrating excellent negative predictive value. This means that when a software deems a mammogram normal, it is highly reliable in ruling out cancer. In France, screening relies on a double-reading system by radiologists. AI could alleviate this workload, which is increasing as the number of radiologists remains stagnant, by reducing the number of mammograms requiring a second review.
The time saved could be reallocated to better patient support, particularly for those identified as having an increased risk of developing cancer. However, AI does not replace radiologists. Software can still generate false positives, flagging abnormalities that are not cancerous. The challenge is to prevent the misuse of these tools, which could undermine women's confidence in organized screening.
At Tenon Hospital, AI is already used in diagnostics, but always under the radiologist's supervision, who must interpret the images first. The initial human reading remains essential, as French radiologists examine the patient and may perform an ultrasound, which AI cannot do. The French screening system, while not perfect, achieves one of the highest detection rates and survival rates in Europe and globally.
Beyond improving the current system, AI opens the door to personalized screening, tailored to each woman's individual risk. Research indicates that AI can assess five-year risk more accurately than most traditional models. Ultimately, this could lead to more frequent screenings for high-risk women and less frequent ones for low-risk women. A major study led by Suzette Delaloge, chair of the board of the National Cancer Institute, will provide insights into the relevance of personalized screening.
The integration of AI also necessitates adapting the training of future doctors. The Faculty of Health at Sorbonne Université has created dedicated modules, and the College of Radiology Teachers of France has incorporated an AI module into its organized screening curriculum. The goal is to foster critical evaluation of algorithms, whose performance varies depending on the situation.
AI's potential extends beyond screening, particularly in histopathology, for a better understanding of tumor heterogeneity and the development of more targeted treatments. While some of these advancements are still in the research phase, they primarily aim to improve patient care while preserving essential medical expertise.
