piNET–An Automated Proliferation Index Calculator Framework for Ki67 Breast Cancer Images

SIMPLE SUMMARY: Approximately 2.1 million women are affected by breast cancer every year. Invasive disease accounts for 80% of breast cancer cases and is the most common and aggressive type of breast cancer. Early diagnosis is the key to survival. Ki67 biomarkers have been shown to be a promising prognostic biomarker in this regard, but manual proliferation index (PI) calculation is time consuming and subject to inter/intra observer variability which reduces clinical utility. Computational pathology tools can aid pathologists to make the diagnostic process more efficient and accurate. With the advent of deep learning, there is great promise that this technology can solve problems that were difficult to tackle in the past, but more work needs to be done to combat the challenge of multi-center datasets. In this work, a novel Ki67 PI calculator based on deep learning is proposed, called piNET, which is shown to be accurate, reliable, and consistent across multi-center datasets. ABSTRACT: In this work, a novel proliferation index (PI) calculator for Ki67 images called piNET is proposed. It is successfully tested on four datasets, from three scanners comprised of patches, tissue microarrays (TMAs) and whole slide images (WSI), representing a diverse multi-centre dataset for evaluating Ki67 quantification. Compared to state-of-the-art methods, piNET consistently performs the best over all datasets with an average PI difference of 5.603%, PI accuracy rate of 86% and correlation coefficient R = 0.927. The success of the system can be attributed to several innovations. Firstly, this tool is built based on deep learning, which can adapt to wide variability of medical images—and it was posed as a detection problem to mimic pathologists’ workflow which improves accuracy and efficiency. Secondly, the system is trained purely on tumor cells, which reduces false positives from non-tumor cells without needing the usual pre-requisite tumor segmentation step for Ki67 quantification. Thirdly, the concept of learning background regions through weak supervision is introduced, by providing the system with ideal and non-ideal (artifact) patches that further reduces false positives. Lastly, a novel hotspot analysis is proposed to allow automated methods to score patches from WSI that contain “significant” activity.