Evaluating Visually Lossless Compression of JPEG XS, JPEG 2000, HEVC and AV1 in Selected Medical Imaging Modalities

The objective of this study is to evaluate the effectiveness of state-of-the-art codecs in compressing selected medical imaging modalities while maintaining visual quality and reducing file sizes. To achieve this, a detailed comparative analysis is conducted comparing the performance of JPEG XS, JPEG 2000, HEVC, and AV1. The analysis takes into consideration compression efficiency, codec complexity, and visual fidelity in the context of medical imaging. Advanced evaluation methods, including the AIC-2 Flicker test, are utilized to determine the visually lossless threshold, which is crucial for preserving diagnostically important details. Additionally, the study explores the potential of crowd-sourcing as a means of assessing the visual quality of compressed medical images. Subjective lab and crowd-sourcing tests reveal varying proportions of correctly identifying the reference images among participants. Furthermore, the study proposes outlier detection methods to improve the reliability of the subjective evaluation and employs kappa analysis to measure the inter-rater agreements. The study analyzes the encoding time taken on a consumer-level CPU, and the results reveal that JPEG XS maintains a fast and consistent speed across different compression levels. The results also indicate that JPEG XS achieves visually lossless performance for diagnostic purposes at 2 BPP, JPEG 2000 at 1.5 BPP, HEVC, and AV1 at 1 BPP.