I recently blogged about the urgent need for easily accessible, large public healthcare datasets here. I’m happy to follow up with a few fantastic examples of relatively open data. Thanks to Sander Boelders for sending some examples to me. If you know any good examples, please share them in the comments below.
But first a very short summary of the initial post. Public datasets have been critical to major progress in AI, and likewise, healthcare data has great potential to unlock advancements in medical AI. But sadly, public healthcare datasets are still rare. Privacy concerns are solvable and IP concerns are usually misguided as progress is likely to be beneficial for all parties. Finally, it’s a moral imperative as the population ultimately deserves to benefit from their own data.
Now for some of the examples I mentioned! Most still require an application process, data-usage agreements, are available only for non-commercial research and some require training. However, they are all wonderful initiatives and a big step in the right direction to publicly available healthcare data.
BraTS, Brain Tumor Segmentation, is an annual brain MRI tumour segmentation competition. Each year, new data and objectives are set, with an overarching aim to develop models that accurately identify and delineate tumour regions, contributing to improved diagnostic tools in neuro-oncology.
Another major initiative related to brain MRI imaging is ATLAS (Anatomical Tracings of Lesions After Stroke). The objective is to accurately segment lesions after stroke, which is critical for rehabilitation.
The CCG, Center for Cancer Genomics, identifies data as a key resource, and one of their goals is to make genomic data available to the research community. One of the main datasets is TCGA which contains “Clinical, biospecimen, molecular characterization, and imaging data for samples from 11,000 patients spanning 33 cancer types”.
Finally, PhysioNet is a platform that provides access to a vast collection of physiological and clinical data. The datasets have varying levels of access. Quoting directly, the policies are:
- Open Access: Accessible by all users, with minimal restrictions on reuse.
- Restricted Access: Accessible by registered users who sign a Data Use Agreement.
- Credentialed Access: Accessible by registered users who complete the credentialing process and sign a Data Use Agreement.