A Two Step Approach for Whole Slide Image Registration

TL;DR

This paper presents a two-step rigid and non-rigid registration method for whole-slide images, evaluated on real-world breast cancer tissue data from the ACROBAT challenge, demonstrating practical applicability.

Contribution

The paper introduces a novel two-step registration approach specifically designed for real-world WSI data, addressing the gap in current methods' performance in practical settings.

Findings

Median 90th percentile registration error is 1,250 micrometers on validation data.

Method performs effectively on real-world breast cancer tissue images.

Addresses the challenge of applying WSI registration in routine diagnostics.

Abstract

Multi-stain whole-slide-image (WSI) registration is an active field of research. It is unclear, however, how the current WSI registration methods would perform on a real-world data set. AutomatiC Registration Of Breast cAncer Tissue (ACROBAT) challenge is held to verify the performance of the current WSI registration methods by using a new dataset that originates from routine diagnostics to assess real-world applicability. In this report, we present our solution for the ACROBAT challenge. We employ a two-step approach including rigid and non-rigid transforms. The experimental results show that the median 90th percentile is 1,250 um for the validation dataset.