Neural Image Compression

About

Contact email:

francesco.ciompi@radboudumc.nl

Version:

47b73f2d-c7e3-417c-b2a7-90097ceaa932

Last updated:

Feb. 5, 2022, 8:09 p.m.

Associated publication:

Tellez D, Litjens G, van der Laak J, Ciompi F. Neural Image Compression for Gigapixel Histopathology Image Analysis. IEEE Trans Pattern Anal Mach Intell. 2021;43(2):567-578.

Inputs:

Generic Medical Image

Outputs:

Neural Image Compression (Compress whole slide)
Neural Image Compression with 90 degree rotation augmentation (Neural Image Compression with 90 degree rotation augmentation)
Neural Image Compression with 180 degree rotation augmentation (Neural Image Compression with 180 degree rotation augmentation)
Neural Image Compression with 270 degree rotation augmentation (Neural Image Compression with 270 degree rotation augmentation)
Neural Image Compression with horizontal flip augmentation (Neural Image Compression with horizontal flip augmentation)
Neural Image Compression with vertical flip augmentation (Neural Image Compression with vertical flip augmentation)
Neural Image Compression with 90 degree rotation and vertical flip augmentation (Neural Image Compression with 90 degree rotation and vertical flip augmentation)
Neural Image Compression with 270 degree rotation and vertical flip augmentation (Neural Image Compression with 270 degree rotation and vertical flip augmentation)

Model Facts

Summary

Compresses Whole Slide Images (WSI) with a convolutional neural network (CNN) described in [1]. A patch of size 128x128x3 is compressed to 128 features. The algorithm saves the compressed slides with additional augmentations (flips and rotations) in the mha format, which can be read with SimpleITK [2].

The WSI must contain a magnification level with 0.5μm pixel spacing (± 0.05).

[1] D. Tellez, D. Hoppener, C. Verhoef, D. Grunhagen, P. Nierop, M. Drozdzal, J. van der Laak, and F. Ciompi,
“Extending unsupervised neural image compression with supervised multitask learning,” in Medical Imaging with Deep
Learning, 2020.
[2] Lowekamp, Bradley Christopher, et al. "The design of SimpleITK." Frontiers in neuroinformatics 7 (2013): 45, https://simpleitk.readthedocs.io

Mechanism

Left empty by the Algorithm Editors

Validation and Performance

Left empty by the Algorithm Editors

Uses and Directions

This algorithm was developed for research purposes only.

Warnings

Left empty by the Algorithm Editors

Common Error Messages

Left empty by the Algorithm Editors

Information on this algorithm has been provided by the Algorithm Editors, following the Model Facts labels guidelines from Sendak, M.P., Gao, M., Brajer, N. et al. Presenting machine learning model information to clinical end users with model facts labels. npj Digit. Med. 3, 41 (2020). 10.1038/s41746-020-0253-3