LVPNet: A Latent-variable-based Prediction-driven End-to-end Framework for Lossless Compression of Medical Images

Chenyue Song,Chen Hui,Qing Lin,Wei Zhang,Siqiao Li,Haiqi Zhu,Shengping Zhang,Zhixuan Li,Shaohui Liu,Feng Jiang,Xiang Li
Harbin Institute of Technology; Nanjing University of Information Science and Technology; Dalian University of Technology; Nanyang Technological University
Medical Image Computing and Computer Assisted Intervention (MICCAI), 2025
PaperCode
Overview of the LVPNet compression pipeline.

LVPNet learns a compact global latent variable, predicts pixel distributions from multi-scale context, and compensates quantization loss for efficient end-to-end lossless medical image compression.

Abstract

Lossless medical image compression must preserve every diagnostic detail while reducing storage and transmission cost. LVPNet combines latent-variable coding with prediction-driven entropy modeling. Its global multi-scale sensing module extracts compact context, a quantization compensation module corrects information lost when the latent representation is discretized, and a probabilistic predictor models the residual image distribution. The unified framework improves compression efficiency across several medical-image datasets.

End-to-End Compression Framework

LVPNet framework with its GMSM and QCM components.

The encoder produces a quantized latent variable for entropy coding. Decoded context is refined by QCM and used to predict a probability distribution for losslessly coding every image pixel.

Multi-Scale Context and Compensation

Feature-map visualization of LVPNet modules.

Feature maps illustrate how the global multi-scale sensing module aggregates broad anatomical structure and how quantization compensation restores residual information.

Sampling-Rate Analysis

Ablation study of latent-variable sampling rate.

The sampling rate controls the trade-off between latent storage overhead and prediction accuracy; the selected operating point gives consistently strong compression across datasets.

BibTeX

@inproceedings{song2025lvpnet,
  title={{LVPNet}: A Latent-variable-based Prediction-driven End-to-end Framework for Lossless Compression of Medical Images},
  author={Song, Chenyue and Hui, Chen and Lin, Qing and Zhang, Wei and Li, Siqiao and Zhu, Haiqi and Zhang, Shengping and Li, Zhixuan and Liu, Shaohui and Jiang, Feng and Li, Xiang},
  booktitle={Medical Image Computing and Computer Assisted Intervention},
  year={2025}
}