Causal Multi-Stream Lossless Compression of Multimodal Volumetric Magnetic Resonance Imaging with Explicit Physical Payload Accounting

Co-registered multimodal magnetic resonance imaging (MRI) contains redundancy within each volume and across an ordered set of contrasts. A lossless codec can exploit this dependence only if a standalone decoder can reproduce every prediction, context query, and payload boundary. We present MM-CLICK, the proper name of the evaluated codec rather than an expandable acronym, as a causal framework for four-volume MRI streams, not as a new predictor or entropy-coding primitive. The codec fixes the FLAIR → T1w → T1gd → T2w serialization, integrates established causal prediction and context-adaptive coding components through decoder-reproducible state updates, regenerates probability contexts from reconstructed samples, and includes boundary data and decoder-required assets in the physical payload. On TEST266 (n = 266), MM-CLICK achieved 1.0877 complete physical bits per source scalar (bpp), compared with 1.0778 bpp for HEVC-RExt; the paired difference was not statistically significant (Wilcoxon p = 0.1873). The measured rate was lower than that of JPEG-LS (1.3626 bpp), reversible JPEG2000 (1.5259 bpp), and BCM-Net (1.3470 bpp) under their stated physical-object scopes. Standalone decoding exactly reconstructed all 266 canonical little-endian uint16 arrays. These results demonstrate a reproducible lossless MRI codec that combines causal multimodal organization, decoder-regenerated contexts, physical payload accounting, and bit-exact verification.