Early Detection of Prostate Cancer with AI
Nearly half of older men will have cancer in 2024, while the percentage for women will be 32.7, according to estimates from the Spanish Cancer Registry Network (REDECAN). This year, the most diagnosed type of cancer in women will continue to be breast cancer (36,395 new cases), and in men, prostate cancer (30,316 new cases). Additionally, the World Health Organization (WHO) ranks prostate cancer as the fourth most common type of cancer globally. In its latest report from 2020, it recorded an incidence of 1.41 million cases. The aging population and increased life expectancy mean the future outlook is not very promising. The report published by The Lancet Commission on prostate cancer predicts an increase in the incidence of this disease, potentially doubling by 2040 (2.9 million new cases).
The Importance of Early Diagnosis of Prostate Cancer
The growth rate of prostate tumors is very slow, making diagnosis difficult. Additionally, this type of cancer typically does not present symptoms until it is in very advanced stages. Almost 1 in 3 cases are detected when metastasis is already present, meaning the cancer cells have spread to other organs. In this situation, the patient’s prognosis is much worse. Furthermore, to avoid underdiagnosis or unnecessary treatments, it is crucial to differentiate between tumor types. Treatment decisions will be based on this differentiation.
Clinically significant prostate cancer (csPCa) involves the presence of cancer cells with the potential to grow and spread. This is the main difference from non-clinically significant cases, where the cells do not exhibit high aggressiveness. Early detection and estimation of the cancer’s grade and stage are crucial to determine the risk and aggressiveness of the tumor and to guide treatments.
Conventional methods for detecting csPCa, such as prostate-specific antigen (PSA) testing or biopsy, have proven useful over the years. However, they are highly invasive and tend to yield false positives and negatives. To reduce the impact of these tests on patients, researchers have explored the use of technologies such as Artificial Intelligence (AI) for diagnostic support in recent years.
How Can AI Help in the Early Diagnosis of Prostate Cancer?
Over the past decade, AI has experienced exponential growth and changed the way we interact with technology across various sectors of society. This evolution is also present in the clinical field, where these tools have revolutionized how diseases are diagnosed, treated, and managed today. AI algorithms allow the identification of hidden patterns and relationships in clinical data. They have already demonstrated high effectiveness in analyzing images, such as X-rays or MRI scans. Additionally, they facilitate the high-precision processing of large volumes of images in seconds, enabling the identification of potential anomalies, such as tumors. This makes them a very useful tool for diagnosing many diseases.
Health Data and Privacy
The General Data Protection Regulation (GDPR) classifies health data as highly sensitive. Therefore, the requirements for conducting studies with this data are very stringent. This slows down the process for research entities to access and use this information to train AI. However, using this information is crucial to advancing technologies for the early detection of prostate cancer.
In this context, since 2023, Gradiant, along with 11 other partners, has been working on the Horizon Europe FLUTE project. We are developing an AI model for predicting csPCa by merging clinical patient data with information extracted from MRI scans. We will train this model federatively at three data centers in Spain, Belgium, and Italy, ensuring data security and privacy through innovative technologies. These data centers are associated with three highly relevant European hospitals (Fundació Hospital Universitari Vall d’Hebron, Centre Hospitalier Universitaire de Liège, and Istituto Romagnolo per lo Studio dei Tumori Dino Amadori). They will provide the clinical knowledge and patient data necessary to implement the AI models.
Federated learning will allow the development of AI algorithms in a decentralized manner, within the hospitals themselves. This avoids sharing patient information, with the advantage that the data will not leave their original locations, thus improving security. At the same time, there will be significant benefits to the algorithms’ results, as they can be trained on a larger amount of data from different locations. This will greatly improve their performance and tumor detection capabilities.
