Optimizing cancer care is one of the major challenges currently facing healthcare systems. In Chile, approximately 60,000 new cases of cancer are diagnosed each year, and roughly 30,000 people die from the disease.
Tumor assessment and monitoring are among the major challenges in oncology, as they require highly demanding processes, such as reviewing images, identifying lesions, and comparing different control images to determine whether a treatment is working. Although this process follows highly reliable standardized protocols, its implementation remains manual, demanding, and variable due to subjectivity.
In light of this situation, an interdisciplinary team of researchers from the University of Chile and the University of Concepción has developed a database of oncology images that aims to become a key component in advancing the use of artificial intelligence for cancer monitoring in Chile.
The study was published in the journal *Scientific Data*, and the research was conducted with the participation of experts from the University of Chile Clinical Hospital, the University of Chile School of Medicine—through the Department of Radiology, the SCIAN-Lab, the Center for Medical Informatics and Telemedicine (CIMT), and the Institute of Biomedical Sciences (ICBM)— and the Faculty of Engineering at the University of Concepción, through its Department of Computer Science and its Center for Data and Artificial Intelligence.
A Contribution to Oncology Through Artificial Intelligence
The database includes 1,246 lesions segmented from 58 CT scans of 22 cancer patients treated at the University of Chile Clinical Hospital, and includes primary tumors, metastases, and lymph nodes, along with other clinical measurements.
One of the study’s main contributions is that it addresses a specific gap in the field: the lack of comprehensive, well-documented clinical data available under responsible open-access conditions for the development of artificial intelligence.
Although partial databases already existed, focusing on specific organs or types of lesions, there was no open, comprehensive resource in this field. “When we started this project, there was no publicly available data covering the entire RECIST process, including measurements and longitudinal tumor follow-up. There were isolated data points, but no comprehensive dataset like this,” notes researcher Constanza Vásquez.
Experts say that the lack of well-annotated data has been one of the main barriers to the advancement of artificial intelligence in local oncological imaging. In this case, moreover, it is not just a matter of collecting images, but of building a database validated by radiology specialists, with lesions delineated one by one and linked to a real clinical protocol. “We are making this data available that did not exist before. So, if someone wants to tackle this problem today, they have direct access to it,” notes Dr. Guillermo Cabrera.
Unlike other partial datasets, this database includes primary tumors, metastases, and lymph nodes, in addition to clinical measurements associated with tumor monitoring. The study provides the first publicly available dataset with comprehensive segmentations of all measurable lesions alongside their RECIST measurements, opening up new possibilities for training models, validating tools, and moving toward more consistent assessments of tumor burden.
Regarding the project’s outlook, which is aimed at moving toward further validation, one of its lead authors, Roberto Rojas, emphasizes: “Generating this data has a huge impact, because such data is scarce and very difficult to collect.” Although the study does not yet provide a model ready for immediate use in hospitals, it does lay a fundamental foundation for adapting future artificial intelligence tools to clinical practice in Chile.
Interdisciplinary collaboration and national reach
The research addresses a fundamental challenge: the underrepresentation of Latin America in the population data used to train medical artificial intelligence. “Artificial intelligence relies heavily on the data used to train it, and the Global South remains underrepresented. That is why it is so important to contribute data that reflects our reality,” explains researcher Steffen Härtel.
Regarding the study’s projections, the team hopes to expand this line of research into a multicenter strategy—ideally one focused on Latin America—that will better reflect the diversity of cases across the continent. “The ideal scenario would be to establish a tumor database in Chile that covers a large portion of the population and addresses our specific needs. It is very important that we be able to characterize this reality at the national level,” concludes Constanza Vásquez.
Read the original article on the University of Chile website.
