ONCOVALUE has developed a unique annotated dataset to train an AI-based imaging tool designed to support the assessment of metastatic breast cancer. With Siemens Healthineers leading its development, the tool ultimately aims to help evaluate which therapies are effective by providing objective, image-based information on disease progression. Using longitudinal medical imaging data (i.e. imaging data over multiple timepoints) these datasets can help train various aspects of imaging AI models to build automated tools for assessing disease progression specifically from CT scans of breast cancer patients with metastases in the lung or liver.
Led by HUS, 235 patients have been annotated—each with multiple scans and multiple lesions, all of which are manually tracked over time by expert radiologists. While the number of patients may seem modest, the depth and quality of annotation are exceptional. As of May 2025:
- 1238 CT scans have been annotated
- Covering 2,701 unique lesions in liver and lungs
- Resulting in 10,230 individual lesion annotations
For each patient, multiple CT scans are contributed, with each scan containing multiple lesions. Many of these lesions have been manually tracked across different timepoints. This means ONCOVALUE doesn’t just capture snapshots of disease—it tracks how metastatic lesions evolve throughout treatment.
These longitudinal annotations—tracking the same lesions across different phases of treatment—are rare in medical imaging research and represent a distinctive strength of the ONCOVALUE project. This enables the development of an AI pipeline—a set of models and algorithms working together to interpret not just individual images, but the full clinical trajectory of each patient.
Enhanced Datasets
To ensure the reliability and consistency of these annotations, the dataset has been thoroughly reviewed and further enhanced through:
- Automatic organ-level localisation of annotated lesions
- Automated temporal consistency checks across timepoints
This annotated dataset powers the development of an AI-based tool for assessing disease progression in longitudinal CT data. Given a longitudinal series of CT images from a breast cancer patient, the AI-based tool detects metastatic lesions, tracks them across timepoints, quantifies their changes, and aggregates the results to assess disease progression, stability, or response to therapy in a standardized manner. Unlike conventional approaches that evaluate a single point in time, ONCOVALUE’s AI tool offers a temporal perspective that enables more robust evaluation of treatment effectiveness across patient populations and supports evidence-based decision making.
Next Steps
While initial work has focused on lung metastases, the pipeline is now being expanded to detect and quantify liver metastases, using the existing annotation base. Plans are underway to involve additional hospitals across Europe to validate the current results and support the expansion of the dataset, further increasing its clinical depth and diversity.
ONCOVALUE is not just developing AI models—it’s laying the groundwork for AI-driven monitoring of how metastatic disease evolves under treatment. By capturing how disease evolves under treatment, the project supports both large-scale evaluations and future applications in precision medicine.
