Diagnosing Breast Cancer With Artificial Intelligence

Breast cancer is the most common cancer in the world today, and one of the deadliest. To make an accurate diagnosis and choose the right treatment, doctors need to analyze tissue samples taken by biopsy. These tissues are then stained and examined under a microscope. It’s a complex process, requiring time and the experienced eye of a specialist, called a pathologist.

To speed up this crucial step, a team of researchers from ÉTS (École de technologie supérieure) and CNRS, in collaboration with two professors from Université Paris-Saclay (CentraleSupélec), are working on an artificial intelligence-based tool (AI). Their aim: automating part of tissue image analyses to support the work of pathologists.

AI Capable of Learning to Recognize Tissues

AI, and more specifically deep learning, is already being used in several areas of medicine. But to work properly, these systems require large quantities of images already analyzed and annotated by experts—which is time-consuming and costly.

In addition, tissue images vary greatly from one hospital to another, depending on the conditions under which they are taken or the staining methods used. These differences can mislead the algorithm.

To work around these problems, researchers rely on what is known as a foundation model: an AI trained to spot common patterns in a large number of varying images. Once properly trained, the model can be adapted to new tasks with very few images—sometimes a single annotated image is all that’s needed.

Another innovation is that, instead of simply pinpointing the areas where the tumour is located, the AI will also be able to take into account physicians’ comments. This will help to better identify the type of tumour, its aggressiveness and other important characteristics.
 

AI Capable of Recognizing Its Limitations

Before being used clinically, AI must be able to recognize when it is not sure of its answer. A wrong prediction can have serious consequences. Researchers are therefore developing strategies to enable the algorithm to signal when it is “in doubt,” so that a human specialist can examine these cases more closely.

A Tool Designed to Help, Not to Replace

The aim is not to replace specialists with machines. It is to save time, so that specialists can analyze more cases and concentrate on the most complex situations. Thanks to this tool, pathologists can be more efficient, while maintaining a high level of accuracy.

Finally, the researchers have chosen to freely share their code and models with the scientific community, thereby accelerating progress in the crucial field of women’s health.