Barcodes for Image Search

About

Background Medical imaging software is estimated to be a $2 billion USD market and projected to spike to $35 billion USD by 2019. This growth is linked to the explosion in imaging data and the need for big data processing and analytics delivered over cloud computing platforms. Within the radiology field, disease misdiagnosis is a major problem that not only leads to increased healthcare costs but also tragically to the premature loss of life. There are estimated to be 500,000 cases of misdiagnosis in the U.S. alone which quite often leads to costly malpractice lawsuits and by extension cost pressure on health insurance premiums for everyone. Many of these errors could be avoided by expanding the radiologists access to a larger database of existing patient images associated with confirmed diagnosis to compare to current patient images and thus enhance the decision making process. Description of the invention Waterloo has developed a method that uses specialized algorithms (based on Radon transforms) to convert images, particularly the region of interest (ROI) in an image (eg. tumour), into a unique barcode representation.  These barcode representations can then be used to annotate images with searchable ROI information (ie. what a confirmed tumour looks like from existing data). This approach is particularly useful for medical imaging whereby any lesion and suspicious mass can be “barcoded” (see figures on the left panel).   As these barcodes are short compact binary codes (0’s and 1’s) they are inherently in a format enabling the fastest level of computational speed which makes them ideal for constructing a highly efficient image search engine. By annotating large publicly available image databases (ie. PACS and RIS) with ROI barcodes, a searchable image database can be created that can be used by radiologists to compare with newly generated patient images.  This approach enables radiologists to quickly to get a “virtual second opinion” of their own diagnostic impression and thus reduce the chances of a misdiagnosis. Advantages Barcodes are compact and thus don’t’ require much memory storage Binary information makes search very fast Both global (similar images) and local (similar image parts) search possible Radon barcodes can be reversible using customized software to reproduce the original image Can be extended to video applications Barcode representations do not represent a privacy problem Potential applications Diagnostic Radiology Pathology Satellite Imaging Internet/personal image search Reference 8810-7411 Patent status U.S. provisional patent filed Stage of development Matlab prototype developed Validated against state-of-the-art using a public benchmark dataset with 15,000 images Seeking industrial partner for medical imaging processing Studies for additional markets are on-going