Our preliminary results

LUNG-PREDICT: Artificial Intelligence for Early Detection of Lung Cancer

Lung cancer is the leading cause of cancer-related death worldwide. One of the main reasons is that diagnosis often occurs too late. Detecting the disease before it progresses radically changes the prognosis; however, current diagnostic pathways rely heavily on imaging tests and invasive procedures that are not always readily available or accessible to all patients.

In this context, the LUNG-PREDICT project, developed by Premium Research in collaboration with IDIBAPS and Hospital Clínic of Barcelona, has explored over the past year whether it is possible to build a clinical decision-support tool based on artificial intelligence, using only data that are already routinely generated in the laboratory.

The approach: non-invasive and accessible

The methodological starting point of the project is based on a practical observation: patients with suspected lung cancer already have, from the outset, blood test results including tumor biomarkers, hematological and biochemical parameters, as well as basic clinical variables. These data exist, are readily available, and until now had not been systematically leveraged as the basis for a predictive model.

The work initially focused on structuring and auditing a patient cohort from the fast-track pathway at Hospital Clínic, verifying data quality and completeness prior to any modeling. Next, the predictive capacity of each variable was analyzed individually, confirming that no single marker is sufficient on its own to reliably stratify risk. This exploratory phase underscores the rationale for a multivariate approach: the relevant information lies not in a single parameter, but in the combination and interaction among them.

On this basis, different machine learning algorithms were trained and compared, applying cross-validation techniques and evaluation on data not previously seen by the model during training. A key aspect of the work has been the incorporation of explainability tools that make it possible to identify which variables contribute most to each individual prediction—an essential requirement for a system to be interpreted and used with confidence by clinical professionals.

The result is a system that, in its current state, functions as a decision-support tool rather than a replacement for clinical judgment. Its purpose is to help prioritize which patients require more urgent evaluation and which present a lower-risk profile, thereby contributing to a more efficient management of the diagnostic pathway.

More results are coming soon. We are also working on scientific publications.