Abstract
Purpose :
We designed the ARI Network Hub: an online platform to share and process swept-source (SS) OCT angiography data. Our goal was to bring together leading clinicians with scientists and developers to accelerate retinal research through collaboration. In this work we analyze the success of this cloud-based solution in the 4 years since origination.
Methods :
We set up a platform where different sites can upload deidentified data directly from a PLEX® Elite 9000 (ZEISS, Dublin, CA) instrument into the ARI Network Hub, and share it in collaboration with other sites as well as execute prototype algorithms. The initial 4-year goal was to add up to 220 sites on board. Prototype algorithm development for innovation was shaped by the clinicians part of this collaborative effort. We tracked number of sites, collaborations started between them, data uploaded, prototype algorithms shared and algorithms executed within the system. We analyzed scalability needs in terms of data storage, processing power and bandwidth to further understand the optimal design of such a cloud-based solution.
Results :
195 sites joined within the 4 years, leading to more than 289 independent publications derived from this collaborative effort. Data uploads were minimal during year 1 but increased during year 2 at a stable rate of ~100 and ~0.7TB per year quarter (/y.qtr) in terms of independent OCT volume datasets and size, respectively. Growth increased ~150%/y.qtr in both number and size in the remaining 2 years (Figure 1). Data uploaded after 4 years was a total of 22000 datasets holding 17TB. 23 processing algorithms were shared across participating sites, with macular density quantification algorithms being of highest usage. Algorithm usage saw an increase with the number of algorithms made available and data shared within the system. There was high non-uniformity across sites in terms of data shared, as well as in algorithm use both across sites and over time, highlighting the need of quick scalability of the design both in storage and calculations.
Conclusions :
The proposed cloud-based solution was adopted and proved to accelerate research, especially considering the streamlined collaboration between institutions and device manufacturers. The vast amount of data acquired and collected represents high potential for future collaborative research and algorithm development.
This is a 2021 ARVO Annual Meeting abstract.