PATH Diagnostic Image Repository
Global rapid diagnostic test image database to support the development of connected diagnostics for low-resource settings.
Access to reliable, complete, and ethically sourced diagnostic data is a key challenge to developing high-quality digital health technologies for low-resource settings. To support and accelerate the development and validation of digital diagnostics, PATH has created a scalable, user-friendly, and secure library of rapid diagnostic test (RDT) images and related metadata. This globally accessible resource can be made available to diagnostic developers demonstrating commercial efforts to support low-and middle-income countries.
Access to this information can enable our partners to develop digital readers and machine learning algorithms that support test result interpretation (i.e., controlled images taken under varying conditions can be used to train computer vision models) and digital connectivity to improve access to and quality of appropriate, point of care diagnostic systems in LMICs.
The PATH Diagnostic Image Repository is housed on a Microsoft SharePoint platform with carefully managed permissions and access controls. Users can access, export, and leverage diagnostic datasets, while maintaining privacy and standards. Currently, it offers images of COVID-19 and HIV RDTs, with additional disease areas and diagnostic technologies under review for continued expansion.
The COVID-19 dataset contains 32,000 images and associated metadata, including a library of 24,000 images that can be used by developers to train machine learning algorithms and around 8,000 images available for algorithm validation purposes. This includes a collection of 32x32 color images of 1,397 unique RDTs, each with 25 different attributes such as defined lighting conditions, sample concentrations, and backgrounds. Images were taken using mobile devices and conditions chosen to reflect the typical environment and quality available in low and middle-income countries, as seen in testing for COVID-19.
The HIV dataset contains 12,000 images, each with an accompanying annotation file containing 12 attributes.
To access the repository, please send a request to firstname.lastname@example.org. User requests will be reviewed on a rolling basis, and global access commitments must be confirmed before access is provided.
Please email email@example.com with questions or to learn how you can donate diagnostic images and metadata to this image repository.