Single Shot AI-assisted quantification of KI-67 proliferation index in breast cancer

TL;DR

This work tackles the variability of Ki-67 proliferation scoring in breast cancer by deploying an AI-assisted approach based on YOLOv8 for automated two-class detection of Ki-67–positive and Ki-67–negative tumor cells in hotspot regions. A dataset of 180 high-resolution Ki-67 IHC images was annotated and expanded to 1,863 samples, with a transfer-learning pipeline leveraging COCO-pretrained weights. Among YOLOv8 variants, the Medium model achieved the best detection performance, with $mAP_{50}$ exceeding $85$ for Ki-67–positive cells and about $73$ for negatives, reflecting both data imbalance and staining clarity. The method offers a scalable, objective alternative to manual scoring and demonstrates potential for integration into pathology workflows, with future work focusing on user-friendly interfaces and multi-institutional validation to improve generalizability and adoption.

Abstract

Reliable quantification of Ki-67, a key proliferation marker in breast cancer, is essential for molecular subtyping and informed treatment planning. Conventional approaches, including visual estimation and manual counting, suffer from interobserver variability and limited reproducibility. This study introduces an AI-assisted method using the YOLOv8 object detection framework for automated Ki-67 scoring. High-resolution digital images (40x magnification) of immunohistochemically stained tumor sections were captured from Ki-67 hotspot regions and manually annotated by a domain expert to distinguish Ki-67-positive and negative tumor cells. The dataset was augmented and divided into training (80%), validation (10%), and testing (10%) subsets. Among the YOLOv8 variants tested, the Medium model achieved the highest performance, with a mean Average Precision at 50% Intersection over Union (mAP50) exceeding 85% for Ki-67-positive cells. The proposed approach offers an efficient, scalable, and objective alternative to conventional scoring methods, supporting greater consistency in Ki-67 evaluation. Future directions include developing user-friendly clinical interfaces and expanding to multi-institutional datasets to enhance generalizability and facilitate broader adoption in diagnostic practice.

Figures (4)

• Figure 1: The Ki67 Detection Workflow using YOLOv8 begins with the data collection of 180 Ki67-labeled IHC images, which are manually annotated by domain experts to label Ki67-positive (red) and Ki67-negative (green) tumor cells. During preprocessing, images are resized to 640 x 640 pixels, normalized, and converted into an appropriate color space to ensure consistency in model training. To enhance dataset diversity, augmentation techniques such as rotation, flipping, scaling, and noise addition are applied, expanding the dataset to 1,863 images and improving model robustness. For model selection, three YOLOv8 variants - Nano, Small, and Medium - are evaluated, with 80$\%$ of the dataset allocated for training and 10$\%$ for validation to optimize model parameters and prevent overfitting. The model's performance is assessed using precision, recall, and mAP50, ensuring reliable detection of Ki67-positive and negative tumor cells. After evaluation, YOLOv8 Medium is identified as the best-performing model, demonstrating superior bounding box regression and detection accuracy. The final model is tested on a separate testing subset, confirming its generalization capability. In the clinical integration phase, the validated model is deployed into pathology workflows for automated Ki67 scoring, reducing manual variability and improving diagnostic consistency. The results are provided with bounding box overlays, allowing pathologists to efficiently assess tumor proliferation, aiding in prognosis and treatment planning.
• Figure 2: Comparative Performance of YOLOv8 Variants for Ki67 Cell Detection
• Figure 3: Increase in the performance of the best YOLOv8 variant.
• Figure 4: Representative images demonstrating the YOLOv8 Medium variant performance in detecting Ki67-positive and negative tumor cells. Red bounding boxes indicate accurate detection of Ki67-positive tumor cells, closely matching ground-truth annotations with minimal false positives. Green bounding boxes represent Ki67-negative tumor cells, showing occasional missed detections or minor localization discrepancies, leading to false negatives, consistent with the quantitative analysis. These visual outputs confirm model proficiency, highlighting clinical reliability for precise Ki67 proliferation index assessment.