NEXTMAP
Harnessing AI for cancer prevention and treatment
Cancer healthcare as an ideal testbed
Prominent challenges in current artificial intelligence (AI) research include the frequent need to rely on noisy and incomplete data from multiple sources, the common presence of blocks of missing data, a call for predictions that are explainable and fair, and the fact that biases in the training data are carried over to future predictions. It is essential to confront and study these challenges in real-world scenarios where the extent of these problems is fully exposed. Cancer healthcare constitutes an ideal domain and testbed due to the intricate complexity of the prediction problem, the vast diversity of data sources and modalities, and the wide range of relevant prediction targets. A common case involves predicting future events from a history of past events.
Our Mission
NEXTMAP will accelerate AI adoption in the public sector by enabling more efficient use of multimodal health data, ensuring robust and trustworthy AI deployment, and lowering implementation costs through reusable frameworks. The project advances state-of-the-art AI methods, tackling challenges in data integration, explainability, and real-world clinical applicability. By integrating AI into medical workflows, NEXTMAP fosters equity in healthcare, enhances national AI competence, and enables personalized patient monitoring and treatment. The system addresses gaps in cancer care by learning from real-world data, including underrepresented populations, improving diagnostics, and streamlining treatment plans. NEXTMAP aims to improve quality of life through earlier disease detection, better treatment outcomes, and reduced healthcare costs. By leveraging Norway’s rich healthcare data and pioneering AI research, the project meets the challenges of an aging population and a shrinking workforce, aligning with global health goals like the UN's Sustainable Development Goal 3: Good Health and Well-Being.
Project themes
(Click on image to see all themes and tasks)
Figure 2: Nextmap themes and tasks.
The center's research is organized into five themes covering the tasks that must be solved for a next-event prediction problem. These themes conceptualize a path from heterogeneous multimodal raw input data to practical applications. In the example shown, the data represent observations from cancer patients, and each patient is represented by several data points reflecting observations taken at different times and representing various modalities. Modality-specific AI agents map each data point to an initial representation (embedding) (Theme 1), and a team of AI fusion agents subsequently map all these initial embeddings into a point in a unified embedding space (Theme 2). A point in this space represents a patient's entire history across all time points and modalities. This high-level representation also embeds information from the scientific literature, relevant clinical studies, etc. Based on the high-level representation, one may develop tools, procedures, and guidelines to enable the development of AI models for specific applications (Theme 3). Finally, one may construct concrete applications such as clinical decision support systems, predictive analytics tools, etc (Theme 4). Applications to other domains than cancer will also be considered (Theme 5).