Artificial intelligence is reshaping oncology, offering unprecedented capabilities in diagnosis, prognosis, and treatment planning. But even the most advanced algorithms cannot solve one of the field’s most pressing problems alone: the lack of diverse, representative, and globally accessible data.
The truth is simple: Cancer does not look the same in every patient, every population, or every hospital.
Yet many of today’s AI models are trained on datasets that reflect only narrow demographics or limited geographies. This is what we call the Remaining84 problem: Currently, most cancer data used to train AI models comes from just 16% of the world’s population. The other 84% remains largely underrepresented. Without global collaboration, AI risks reinforcing inequities in cancer care rather than eliminating them.
Why Global Collaboration Matters
The potential of AI in oncology depends on genetic, technological, and institutional diversity. Collaboration across borders brings:
-
Data diversity: Ensuring AI models reflect cancers across different populations, ethnicities, and regions.
-
Technical interoperability: Harmonizing imaging, pathology, and omics pipelines across institutions.
-
Shared validation: Allowing models to be tested in varied clinical environments, ensuring robustness and fairness.
-
Resource equity: Bringing advanced diagnostics not just to high-resource centers, but also to hospitals and clinics worldwide.
Without such efforts, AI will remain a tool for a few, rather than a catalyst for global oncology progress.
Lessons from Recent Developments
Over the past three years, leading journals have emphasized that the most impactful AI studies come from multi-institutional, cross-continental collaborations. Whether in breast cancer screening, pathology foundation models, or multimodal data integration, progress accelerates when researchers and clinicians align across borders.
But collaboration is not just a scientific requirement; it is also an ethical one. In oncology, where outcomes differ dramatically between high- and low-resource settings, AI must be developed with inclusivity at its core.
PAICON’s Perspective
At PAICON, we believe that collaboration is not optional; it is the foundation of sustainable innovation. By building partnerships with hospitals, research institutions, and industry leaders worldwide, we are creating a genetically and technologically diverse datalake that supports fair, transparent, and robust AI solutions.
We have brought this commitment to life through the Remaining84 initiative, PAICON’s call to close the gap in cancer AI. We do not just talk about diversity; we show it. On PAICON, you can explore our interactive global map that highlights exactly where our datasets originate. This level of transparency is rare in cancer AI and underscores our belief that true innovation requires both accountability and inclusivity. By opening our data story to the world, we demonstrate why PAICON is at the forefront of building AI that works for all patients, not just a privileged few.
Our mission is to ensure that the benefits of cancer AI extend beyond select populations and reach patients everywhere. We see a future where foundation models, explainable AI, and multimodal integration are only as strong as the global networks that sustain them.
A Call to Action
The oncology community stands at a crossroads. The next generation of AI will not be defined only by algorithms but by how we collaborate to build them. Together, we can ensure that AI does not just advance cancer care, but also it democratizes it.
Learn more about our commitment to diversity and equity in cancer AI through our Remaining84 initiative.
References
-
Esteva A, Topol EJ. A guide to deep learning in healthcare. Nat Med. 2019;25(1):24–9. doi:10.1038/s41591-018-0316-z.
-
Stenzinger A, Alberter C, Allgäuer M, et al. Artificial intelligence and pathology: from principles to practice and future applications in oncology. Mol Oncol. 2021;15(9):2339-55. doi:10.1002/1878-0261.12913.
-
Mobadersany P, Yousefi S, Amgad M, et al. Predicting cancer outcomes from histology and genomics using convolutional networks. Proc Natl Acad Sci U S A. 2018;115(13):E2970–9. doi:10.1073/pnas.1717139115.
