Optimising Region of Interest Registration for Multiple-Tissue Whole Slide Images

Digital pathology transforms clinical workflows by enabling the analysis of whole slide images (WSIs) on computers. While most methods focus on haematoxylin and eosin (H&E) stained WSIs, immunohistochemistry (IHC) is crucial for biomarker analysis, particularly for assessing tumour-infiltrating lymphocyte (TIL) subtypes inside a region of interest (ROI) selected by a pathologist using Salgado’s criteria. This study first investigates the naive approach, which verifies a state-of-the-art WSI registration method’s robustness for registering single-tissue H&E and multiple-tissue IHC WSIs. Then, to simplify this first attempt by accomplishing registration between single tissues, the study proposes two approaches: splitting the multiple-tissue IHC WSIs, which is considered the baseline, and a virtualised splitting with an incremental resolution optimisation-based technique. ROI registration predictions for TIL assessment will be assessed on IHC WSIs using standard metrics and one derived from a standard landmark-based metric popular in the image registration field. Existing image detection-inspired metrics for evaluating ROI proposals will be proposed and tuned to consider the global viewpoint, where the ROI proposals lie. This study aims to establish a reliable and time-efficient ROI registration procedure for WSIs with multiple stained tissues. This method would enable efficient selection of the ROI from the H&E WSI and potentially reduce the need for pathologist intervention through automatic quality control.