The development of a clinically tested visually lossless Image compression system for capsule endoscopy
Background and objective: Capsule endoscopy investigations currently capture a large number of images that must be processed and transferred inside the capsule. This demands a compression system that is capable of reducing the amount of data processed, encoded and transmitted. In comparison to a lossless system, a near-lossless image compression approach would have lower complexity and reduce the amount of data processed inside the capsule. However, the visual quality of the images compressed using a near-lossless approach must be deemed suitable for diagnostic use by clinicians. Methods: This paper outlines the development and clinical validation of a near-lossless image compression approach based on a Simplified Structure Conversion method, an Optimized Difference Approach, and a Golomb-Rice encoder. Eighteen clinicians participated to assess the suitability of processed videos and images for diagnostic purposes. Results: Our near-lossless approach returned a Peak Signal to Noise Ratio (PSNR) of 43.89182 dB and Structural Similarity measure of 0.996746 and an overall compression ratio (CR) of 5:1. Clinicians assessed the videos compressed by the developed system and rated 82% of the videos as acceptable for diagnostic purposes. Clinicians also compared the images before and after processing with the developed system and rated 86% of the images as "no difference". Conclusion: The high values of the MSSIM, PSNR, CR and the high acceptance of clinicians provided a reliable indication of the high performance of the developed compression system. This matches the image quality requirements of capsule endoscopy for real-world medical diagnosis.