# Revisiting Ki-67 Assessment in Canine Mast Cell Tumours: From Manual Hotspot to Automated Global Analysis

## TL;DR

This paper introduces a new automated method to assess Ki-67 in canine mast cell tumours, improving reliability and potential for predicting outcomes.

## Contribution

A semi-automated workflow using digital pathology and deep learning for global Ki-67 assessment in canine MCTs is developed and validated.

## Key findings

- Global Ki-67 levels strongly correlate with manual hotspot counts and show potential for predicting survival.
- ROI definition required manual adjustments, but Ki-67 quantification was rapid and automated.
- Ki-67-positive cell density in hotspots was up to 38-fold higher than in whole tumour sections.

## Abstract

Canine mast cell tumours (MCTs) show highly variable behaviour, making it difficult to predict outcomes. Ki-67 is a protein that marks all proliferating cells and is commonly used as a prognostic indicator. Traditionally, Ki-67 is measured manually in small “hotspot” areas of the tumour, but results can vary between pathologists, limiting reproducibility. This study developed a semi-automated workflow to measure Ki-67 across the whole tumour tissue section using digital pathology and deep-learning tools. A total of 309 tumours were analysed, and the results were compared with manual hotspot counts, tumour grade, and clinical outcome. The workflow enabled rapid and reliable Ki-67 measurement across a range of hundreds of thousands to millions of cells, though defining the tumour region often required some manual correction. Global Ki-67 levels strongly correlated with manual hotspot counts and showed promising potential to predict survival, though larger studies are needed. Importantly, this approach can be applied to Ki-67 and other biomarkers, providing a standardized, reproducible, and comprehensive method for assessing tumour behaviour and improving prognostic and treatment decisions in veterinary oncology.

Canine mast cell tumours (MCTs) show highly variable behaviour, and Ki-67 is an established prognostic indicator. Conventional Ki-67 assessment is manual and restricted to small hotspot areas, limiting reliability. This study presents a semi-automated whole-tumour tissue section (global) Ki-67 analysis workflow, outlines its limitations, and examines correlations with hotspot counts and clinical outcome. A total of 309 canine MCTs were assessed using a deep-learning-assisted quantification with commercial software. Global Ki-67 metrics were correlated with hotspot Ki-67 counts and histomorphologic tumour grades, as supported by clinical follow-up data from 68 dogs. The defined analytic workflow enabled an overall feasible global Ki-67 assessment in canine MCTs. The region-of-interest (ROI) definition required frequent manual adjustments, whereas Ki-67 quantification was fully automated and rapid. Global Ki-67 metrics correlated with manual hotspot counts, with Ki-67-positive cell density on average twice as high in tumour hotspots compared with whole tumour sections, with differences ranging up to 38-fold. Exploratory survival analyses suggested promising predictive power, warranting validation in a robust survival study. With established digital pathology tools, global whole-tumour assessment of Ki-67 and other biomarkers is feasible. It should become the new standard for defining robust prognostic and predictive markers in canine mast cell and other tumours.