JoVE - Journal of Visualized Experiments
JoVE Visualize
Contact Us

RegWSI: Whole slide image registration using combined deep feature- and intensity-based methods: Winner of the ACROBAT 2023 challenge

  • 0Institute of Informatics, University of Applied Sciences Western Switzerland, Sierre, Switzerland; Department of Measurement and Electronics, AGH University of Kraków, Krakow, Poland.
Computer Methods and Programs in Biomedicine +

|

April 24, 2024

|

April 24, 2024

View abstract on PubMed

Summary

This summary is machine-generated.

This study introduces a novel two-step hybrid method for automatic whole slide image (WSI) registration, achieving high accuracy and efficiency for digital pathology applications without dataset-specific fine-tuning.

Area Of Science

  • Digital Pathology
  • Medical Image Analysis
  • Computational Biology

Background

  • Accurate registration of differently stained whole slide images (WSIs) is vital for diagnosis and prognosis.
  • Challenges include variations in slide preparation, staining, and tissue deformation.
  • Efficient annotation transfer between serial WSIs can significantly reduce time and cost.

Purpose Of The Study

  • To develop a robust, efficient, and accurate automatic registration method for WSIs.
  • To address the limitations of current registration techniques in digital pathology.
  • To provide a generalizable solution applicable to various tissue types and stains.

Main Methods

  • A two-step hybrid approach combining deep learning/feature-based initial alignment and intensity-based nonrigid registration with instance optimization.
  • The method requires no fine-tuning and is directly applicable to diverse datasets.
  • Implemented within the DeeperHistReg framework for versatile WSI processing.

Main Results

  • Achieved 1st place in the ACROBAT 2023 challenge.
  • Demonstrated superior performance compared to state-of-the-art methods on ANHIR, ACROBAT, and HyReCo datasets.
  • Showcased high accuracy in cell-level registration for restained slides and robust performance across different resolutions.

Conclusions

  • The developed method offers automatic, robust, and highly accurate WSI registration.
  • It is a generalizable, out-of-the-box solution for microscopic image registration.
  • The open-source release and reproducible results advance digital pathology capabilities.

Keywords: