Automated and Manual Quantification of Tumour Cellularity in Digital Slides for Tumour Burden Assessment

Abstract

Aims:The residual cancer burden index is an important quantitative measure used for assessing treatment response following neoadjuvant therapy for breast cancer. It has shown to be predictive of overall survival and is composed of two key metrics: qualitative assessment of lymph nodes and the percentage of invasive or in-situ tumour cellularity (TC) in the tumour bed (TB). Currently, TC is assessed through eye-balling of routine histopathology slides estimating the proportion of tumour cells within the TB. With the advances in production of digitized slides and increasing availability of slide scanners in pathology laboratories, there is potential to measure TC using automated algorithms with greater precision and accuracy. Methods: We describe two methods for automated TC scoring: 1) a traditional approach to image analysis development whereby we mimic the pathologists' workflow, and 2) a recent development in artificial intelligence in which features are learned automatically in deep neural networks using image data alone. Results: We show strong agreements between automated and manual analysis of digital slides. Agreements between our trained deep neural networks and experts in this study (0.82) approach the inter-rater agreements between pathologists (0.89). We also reveal properties that are captured when we apply deep neural network to whole slide images, and discuss the potential of using such visualisations to improve upon TC assessment in the future. Conclusions: TC scoring can be successfully automated by leveraging recent advancements in artificial intelligence, thereby alleviating the burden of manual analysis.

Digital Pathology: Advantages, Limitations and Emerging Perspectives

“…Further auxiliary, mostly AI-powered applications, have either already been integrated into commercially available platforms or have been published to be suitable for such integration. Use cases include immunohistochemical Ki-67 evaluation to determine a tumor’s proliferation rate [ 23 ], challenging to quantify biomarkers (e.g., immunohistochemical PD-L1 staining [ 24 , 25 ]), evaluation of residual cancer burden after chemotherapy [ 26 ], or cancer detection and classification algorithms (e.g., in prostate cancer [ 27 , 28 , 29 ]). In a further step towards automation, reports may then be composed using AI-based speech-recognition.…”

“…These applications have variably been included in DP viewers for easy operation. Also, an application to estimate residual tumor cell content after chemotherapy has been developed [ 26 ]. A more complex application for routine diagnostics has been made available as a CE-IVD approved software for the Philips IntelliSite Pathology Solution in 2018.…”

Direct Cellularity Estimation on Breast Cancer Histopathology Images Using Transfer Learning

“…Unlike the previous approach proposed by Peikari et al, our methods require for training only the cellularity labels on image patches instead of the annotations on individual nucleus. The following contributions are made:Our method can directly estimate cellularity from breast cancer slide patches and avoid the segmentation and classification of nuclei, which is similar to the approach in [23]. We validate the transferability to tissue microscopy of the deep features learned from natural images.…”

“…As for automatic cancer cellularity assessment, there are two literatures addressing this challenge [22, 23]. In [22], Peikari et al use a two-stage method consisting of nuclei segmentation and classification.…”

“…This procedure is analogous to the workflow of the pathologist; however, it is subject to the accuracy of segmentation and classification. In [23], Akbar et al fine-tuned two modified Inception [27] networks independently to identify the images as cancerous or noncancerous and predict the cellularity for those cancerous patches. The method is validated using the same dataset as in [22], and superior performance is achieved.…”