The Catalan Institute of Health (ICS) has developed four proprietary artificial intelligence algorithms that have improved the diagnosis of breast cancer. It is a first step in a digital transformation project for pathological anatomy services that will be extended to other tumors. Specialists are already working on an algorithm for lung cancer.
The idea is for AI to help make the diagnosis. In the case of breast cancer, it is based on the quantification of four biomarkers or molecules whose presence indicates a disease process: HER2, Ki67, estrogen receptors and progesterone receptors. The technology has already been used in nearly 600 possible breast tumors throughout the ICS hospital network.
The advances are part of the Digipathics project, which the ICS began deploying in 2021. Since then, more than two million slides with histological preparations (microscopic samples of biological tissues) have been digitized, which form the basis of data that feeds the learning of artificial intelligence algorithms to refine diagnoses.
Pathologists and AI specialists from the Universitat Politècnica de Catalunya collaborate in the development. “Algorithms help to quantify prognostic markers – an analysis that until now was done manually – with larger numbers, faster and more reproducibly, not so dependent on the expertise of who is looking at the images”, explains Jordi Temprana, assistant doctor at the pathological anatomy service of Vall d’Hebron.
This Barcelona center is one of the seven public hospitals that contribute samples to the database, with Arnau de Vilanova, Doctor Trueta, Germans Trias i Pujol, Joan XXIII, Bellvitge and Verge de la Cinta. It is the most important pathological anatomy network in Europe, in which 168 pathologists can share cases, diagnoses and knowledge in real time.
Hospitals produce one million slides (preparation of patient tissue samples) a year and 24 precision scanners work to digitize all samples. Network work and the greater precision provided by digitization means improving speed, security, efficiency and equity in the territory, according to Temprana. In addition, there is better availability of cases for the training of resident doctors. “Algorithms behave as if you had a super-expert always available to make a determination. You can share cases, consult a difficult case more easily and make a faster and more accurate diagnosis”, says Temprana.
So why do we need doctors? “It is not the algorithm that makes the diagnosis, but the doctor, but it helps him. The diagnosis is made by the pathologist by looking at the image, interpreting the image, and the machine provides extra information that can make it faster and safer.”
The application of AI to cancer diagnosis is in its infancy, but it is the future. And they will also extend to the treatment. “We don’t know where the limit is, we don’t know when, but we will see changes in the way we work”, says Temprana: “We started with breast cancer because it is very prevalent, has a lot of social impact and has known biomarkers”.
The ICS network of hospitals is already working to also develop algorithms for the diagnosis of lung cancer, starting this year, as part of the Digipathics project for the digital transformation of pathological anatomy services. Both histological samples (fragments of tissue) and diagnostic-quality images are digitized, which allows the replacement of microscopes with high-resolution screens and facilitates the sharing of high-resolution images by all hospitals in the network. The great challenge of the project consists in the handling and processing of an enormous volume of information, now encrypted – and rising – at one petabyte per year, the equivalent of around one billion books.
