Search

Endometrial cancer (EC) has four molecular subtypes with strong prognostic value and therapeutic implications. The most common subtype (NSMP; No Specific Molecular Profile) is assigned after exclusion of the defining features of the other three molecular subtypes and includes patients with heterogeneous clinical outcomes. In this study, we employed artificial intelligence (AI)-powered histopathology image analysis to identify a novel sub-group of NSMP EC patients that had markedly inferior progression free and disease free survival in a discovery cohort of 368 patients and an independent validation cohort of 290 patients from another center. Shallow whole genome sequencing revealed a higher burden of copy number abnormalities in the identified group, compared to other NSMP EC, in our discovery and validation cohorts. Taken together, our work demonstrates the power of AI to discover new knowledge, identifying a prognostically relevant subset of EC that is unrecognizable with conventional histopathological assessment, refining image-based tumor classification.

The color variation of hematoxylin and eosin (HE)-stained tissues has presented a challenge for applications of artificial intelligence (AI) in digital pathology. Many color normalization algorithms have been developed in recent years in order to reduce the color variation between HE images. However, previous efforts in benchmarking these algorithms have produced conflicting results and none have sufficiently assessed the efficacy of the various color normalization methods for improving diagnostic performance of AI systems. In this study, we systematically investigated eight color normalization algorithms for AI-based classification of HE-stained histopathology slides, in the context of using images both from one center and from multiple centers. Our results show that color normalization does not consistently improve classification performance when both training and testing data are from a single center. However, using four multi-center datasets of two cancer types (ovarian and pleural) and objective functions, we show that color normalization can significantly improve the classification accuracy of images from external datasets (ovarian cancer: 0.25 AUC increase, p = 1.6 e-05; pleural cancer: 0.21 AUC increase, p = 1.4 e-10). Furthermore, we introduce a novel augmentation strategy by mixing color-normalized images using three easily accessible algorithms that consistently improves the diagnosis of test images from external centers, even when the individual normalization methods had varied results. We anticipate our study to be a starting point for reliable use of color normalization to improve AI-based, digital pathology-empowered diagnosis of cancers sourced from multiple centers. © 2021 The Pathological Society of Great Britain and Ireland. Published by John WileySons, Ltd.

Ovarian carcinoma has the highest mortality of all female reproductive cancers and current treatment has become histotype-specific. Pathologists diagnose five common histotypes by microscopic examination, however, histotype determination is not straightforward, with only moderate interobserver agreement between general pathologists (Cohen's kappa 0.54-0.67). We hypothesized that machine learning (ML)-based image classification models may be able to recognize ovarian carcinoma histotype sufficiently well that they could aid pathologists in diagnosis. We trained four different artificial intelligence (AI) algorithms based on deep convolutional neural networks to automatically classify hematoxylin and eosin-stained whole slide images. Performance was assessed through cross-validation on the training set (948 slides corresponding to 485 patients), and on an independent test set of 60 patients from another institution. The best-performing model achieved a diagnostic concordance of 81.38% (Cohen's kappa of 0.7378) in our training set, and 80.97% concordance (Cohen's kappa 0.7547) on the external dataset. Eight cases misclassified by ML in the external set were reviewed by two subspecialty pathologists blinded to the diagnoses, molecular and immunophenotype data, and ML-based predictions. Interestingly, in 4 of 8 cases from the external dataset, the expert review pathologists rendered diagnoses, based on blind review of the whole section slides classified by AI, that were in agreement with AI rather than the integrated reference diagnosis. The performance characteristics of our classifiers indicate potential for improved diagnostic performance if used as an adjunct to conventional histopathology.