18%. This is the rate of additional cancers detected by artificial intelligence compared to traditional methods. This is without increasing the number of false positives.
At the origin of this result, a German work, the PRAIM study. The largest ever achieved in this field. Researchers from the University of Lübeck and the University Medical Center of Schleswig-Holstein analyzed the mammograms of more than 460,000 women between 2021 and 2023.
In detail, all participants had their images examined independently by two radiologists (this is the traditional German method). However, for half of them, at least one of the experts used an AI tool to assist them.
« We simply hoped to prove that AI was as effective as radiologists », Says Professor Alexander Katalinic who led this research. “ The results exceeded all our expectations. »
The study reveals that AI made it possible to identify 6.7 cases of breast cancer per 1,000 women examined, compared to 5.7 cases per 1,000 using traditional methods. This is equivalent to one additional case of cancer detected per 1,000 women.
Beyond improving diagnosis, this technology could considerably lighten the workload of radiologists. In Germany, where 24 million mammogram images are analyzed each year, AI could reduce the number of exams requiring human intervention by 15%.
Hope in other female cancers
But the use of artificial intelligence is not reserved for breast tumors. Coincidentally, at the same time as the publication of the PRAIM study, another work was published, this time Swedish and focused on ovarian cancer.
This study conducted by the Karolinska Institute once again demonstrates that AI can outperform human experts. Analysis of 17,000 ultrasound images from 3,652 patients in 20 hospitals in eight countries showed an accuracy rate of 86.3% for AI, compared to 82.6% for experts and 77.7% for examiners less experienced. AI could thus compensate for the lack of ultrasound experts, reduce unnecessary interventions and speed up diagnoses, particularly in complex cases.
Source : https://www.nature.com/articles/s41591-024-03408-6 – https://www.nature.com/articles/s41591-024-03329-4
