I-Corps: Scalable Storage of Whole Slide Images and Fast Retrieval of Tiles for Next-Generation Image Analytics

点击翻译按钮获取中文摘要

The broader impact/commercial potential of this I-Corps project is pivoted around the emerging whole slide imaging market segment in digital pathology. The digital pathology market is predicted to reach over $800 million by 2025. The motivation of this project stems from the gigabyte size of whole slide images (WSIs), which are digital images of glass slides produced at near optical resolution. With rapid increase in the number of WSIs produced by hospitals and pathology labs, the storage and management of WSIs has become an urgent problem to tackle for next-generation image analytics. The commercial viability of the project can significantly impact researchers, medical professionals, software developers, and IT staff who wish to manage large number of WSIs using modern cluster computing and big data techniques. Thus, next-generation image analytics (e.g., using deep learning) for automatic detection and analysis of cellular and morphological features in human tissues can be performed faster on large numbers of WSIs. The potential societal benefit of the project includes enabling improved diagnosis of diseases by pathologists using next-generation image analytics and the creation of a tech startup leading to new jobs. This project will provide training to two Ph.D. students from underrepresented groups in STEM.This I-Corps project is based on a software technology that aims to solve the fundamental problem of scalable storage of WSIs and fast retrieval of tiles using a commodity cluster and big data techniques. The value proposition of the technology is efficient and cost-effective storage of large-scale WSIs and fast retrieval of tiles to enable next-generation image analytics for human disease diagnosis. The technology encompasses intelligent data partitioning using space-filling curves, in-memory data structures, and effective organization of tiles to enable fast retrieval of tiles during image analysis. It employs space-efficient storage formats to maximize storage efficiency. On an average, it required a few seconds to retrieve a single tile on 80 WSIs using a 16-node cluster. Therefore, we believe next-generation image analytics on WSIs (e.g., using deep learning) can run faster on large number of WSIs, which can consume terabytes of storage, through faster access of image tiles. As the technology relies on commodity hardware and open source software, it is cost-effective and can be easily deployed as a product or a service. The technology has the potential to advance the state-of-the-art in WSI storage and management using a big data approach.This award reflects NSF's statutory mission and has been deemed worthy of support through evaluation using the Foundation's intellectual merit and broader impacts review criteria.