Glandular Segmentation of Prostate Cancer: An Illustration of How the Choice of Histopathological Stain Is One Key to Success for Computational Pathology

Abstract

Digital pathology offers the potential for computer-aided diagnosis, significantly reducing the pathologists' workload and paving the way for accurate prognostication with reduced inter-and intra-observer variations. But successful computer-based analysis requires careful tissue preparation and image acquisition to keep color and intensity variations to a minimum. While the human eye may recognize prostate glands with significant color and intensity variations, a computer algorithm may fail under such conditions. Since malignancy grading of prostate tissue according to Gleason or to the International Society of Urological Pathology (ISUP) grading system is based on architectural growth patterns of prostatic carcinoma, automatic methods must rely on accurate identification of the prostate glands. But due to poor color differentiation between stroma and epithelium from the common stain hematoxylin-eosin, no method is yet able to segment all types of glands, making automatic prognostication hard to attain. We address the effect of tissue preparation on glandular segmentation with an alternative stain, Picrosirius red-hematoxylin, which clearly delineates the stromal boundaries, and couple this stain with a color decomposition that removes intensity variation. In this paper we propose a segmentation algorithm that uses image analysis techniques based on mathematical morphology and that can successfully determine the glandular boundaries. Accurate determination of the stromal and glandular morphology enables the identification of the architectural pattern that determine the malignancy grade and classify each gland into its appropriate Gleason grade or ISUP Grade Group. Segmentation of prostate tissue with the new stain and decomposition method has been successfully tested on more than 11000 objects including well-formed glands (Gleason grade 3), cribriform and fine caliber glands (grade 4), and single cells (grade 5) glands.

Zebrafish xenograft model for studying mechanism and treatment of non-small cell lung cancer brain metastasis

“…2 A). Hematoxylin-eosin (HE) staining mainly distinguishes normal and abnormal cells by the morphology and size of the nucleus [ 31 ]. H1975 cells could be found in the brain of zebrafish at almost single cell resolution (white arrows).…”

Automated identification of glomeruli and synchronised review of special stains in renal biopsies by machine learning and slide registration: a cross‐institutional study

“…One study of prostate neoplasia identification achieved favourable results by requiring that all specimens be prepared and stained by the exact same method. 11 It is intuitive that strict control over a standardised preparation process would allow for optimisation of any CNN-created model. In the absence of such a standard, more extensive training on more samples of differing types will be required.…”

Artificial Intelligence-Enabled Prostate Cancer Diagnosis and Prognosis: Current State and Future Implications

“…Therefore, the reliability of AI algorithms is questionable for other benign or malignant conditions encountered for which the AI algorithm was not trained. 52,53 This can lead to missed diagnosis. Human supervision can resolve this issue.…”

“…Such factors include variations in the image quality, staining methods, staining times, types of scanners used for digitization, and file formats of images, which could affect the performance of AI tools. [9][10][11][12][43][44][45][46][47][48][49][50][51][52][53][54] Differences in the performance of AI algorithms in the test sets as compared with training sets in different settings, such as internal versus external sets, are the proofs of such effects. In a recent study by Eloy et al, 36 scanning error was the single most important identified in the retrospective review of discrepant cases of prostate cancer on biopsies.…”