Over the past decades, digital medical imaging has become ubiquitous throughout the world. Every day, millions of patients are imaged for many clinical reasons, using modalities such as magnetic resonance imaging (MRI), computed tomography (CT), positron emission tomography (PET), digital mammography and digital X-ray systems. Many of these modalities acquire detailed anatomical, structural, and functional data useful for diagnosing, treating, and monitoring diseases in individual patients. These vast amounts of data are archived around the world and represent an invaluable asset for advancing our collective understanding of human biology and for discovering, developing, and validating therapies to treat disease.
While new approaches are now being developed and deployed to increase the utilization of large data sets (e.g. sophisticated technologies to mine "big data"), these advances largely ignore core biomedical image data. A single MRI or CT scan of the human brain may yield thousands of complex images; such images have proven resistant to automated analytical methods due to limited accuracy and the need for manual review and correction. As a result, interpreting complex biomedical images remains a labor-intensive task for human experts (radiologists and other physicians or scientists). So while our enormous and growing archives of biomedical image data play roles in the care of individual patients (e.g. prior film comparisons to assess changes), this resource has been barely touched for more ambitious purposes.
We are attacking this problem from a different angle. Modern information systems have facilitated new ways of mobilizing groups of people with common interests, including "crowdsourcing" strategies to address complex tasks. Our approach can be seen as an advanced variant of crowdsourcing that can have far-reaching educational and research potential.