This is a world first! An AI algorithm has been clinically validated in the detection of breast cancer for a wide variety of specific subtypes.
The study, published in the journal npj Breast Cancer of Nature, is the result of a collaboration between Institut Curie in France and Ibex Medical Analytics, the leader in AI-assisted cancer diagnosis. She is the first to present an algorithm capable of accurately detecting many pathological characteristics in breast biopsies (or tissue samples).
Fast and objective help for pathologists
With more than 2.2 million new cases in 2020, the cancer be you is the most common malignant disease in the world. Usually, the diagnosis of this type of cancer is established following a histological examination of breast biopsy samples, which can be laborious, subjective and therefore a source of error. Furthermore, the number of pathologists has decreased in recent years, while the overall incidence of breast cancer has increased. ” There is therefore a growing need for automated solutions and decision support tools for pathologists to detect cancers faster with the highest accuracy. », Reports a statement of the Curie Institute. The system was developed to intervene in support of the medical profession.
“I was impressed with the results of the study, the very high levels of accuracy and the breadth of detection capabilities offered by the AI technology”
Concretely, the AI algorithm was trained to identify 51 breast characteristics using methods of deep learning, on hundreds of thousands of sample images. Next, the search focused on 841 whole slide images from breast biopsies. The AI algorithm analyzed these images and the results were assessed once morest a consensus diagnosis made by two pathologists. ” I was impressed with the results of the study, the very high levels of accuracy and the breadth of detection capabilities offered by the AI technologysimilar to those of expert pathologists “said Stuart Schnitt, physician, professor of pathology at the Harvard Medical School and co-author of the study.
Distinguish a wide variety of types of breast cancer
Different types of breast cancer have been identified by AI, including rare subtypes. For example, the system distinguished invasive breast carcinoma of non-specific type from invasive/invasive lobular carcinoma, which represent 70% and 10-15% of all cases, respectively. breast tumors invasive. But the AI algorithm also found rare types like metaplastic or mucinous carcinomas. In addition, it has made it possible to identify tumor-infiltrated lymphocytes (TILs), biomarkers of lymphocyte-predominant breast cancers.
« We are delighted with this collaboration, which has enabled certain pathologists from the Institut Curie to gain direct experience of this AI tool. said Anne Vincent-Salomon, doctor and head of the pathology department at the Institut Curie and professor at the University of Paris-Sciences et Lettres. ” TO In the long term, this would allow us to optimize diagnoses, accelerate therapeutic decisions and, ultimately, improve the care of our patients. »