Automated Ki-67 proliferation scoring from histopathology images using Mobile and Cloud technology
Miranda McConnell, Richard Gault, Stephanie G. Craig, David Cutting, Austen Rainer, Jacqueline A. James · Research Portal (Queen's University Belfast) · 2021
The Ki-67 protein is associated with cell proliferation and is a clinical marker for breast cancer tumour aggressiveness. The percentage of immunopositive cells present in a histological image stained for Ki-67 expression informs the proliferation index quantifying tumour aggressiveness. This calculation is frequently carried out through manual assessment that is time consuming and susceptible to human error. Automated image analysis tools for Ki-67 breast cancer images may have a significant impact if they could be integrated in to clinical and digital pathology workflows by reducing workload for pathologists, as well as improving efficiency and accuracy. This work presents the development of a deep learning based model for automated calculation of Ki-67 proliferation scores from stained histological images. The resulting computational model predicts cell types (immunopositive vs immunonegative) with 96% accuracy, the Ki-67 index category with 88% accuracy and the Ki-67 index with lower RMSE than the state of the art models. The predicted mask from the model provides a transparent explanation of the computational decision making. Moreover, the computational model is hosted on a cloud platform and can be utilised through a mobile application designed for this investigation. The proof-of-concept mobile application has the potential to make an impact in many communities, especially in low and middle income countries where there are currently insufficient resources, namely a lack of expensive digital scanners, to support digital pathology in the fields of medicine and education.