Computational segmentation and classification of diabetic glomerulosclerosis

Brandon Ginley, Brendon Lutnick, Kuang Yu Jen, Agnes B. Fogo, Sanjay Jain, Avi Rosenberg, Vighnesh Walavalkar, Gregory Wilding, John E. Tomaszewski, Rabi Yacoub, Giovanni Maria Rossi, Pinaki Sarder

Research output: Contribution to journalArticle

Abstract

Background Pathologists use visual classification of glomerular lesions to assess samples from patients with diabetic nephropathy (DN). The results may vary among pathologists. Digital algorithms may reduce this variability and provide more consistent image structure interpretation. Methods We developed a digital pipeline to classify renal biopsies from patients with DN. We combined traditional image analysis with modern machine learning to efficiently capture important structures, minimize manual effort and supervision, and enforce biologic prior information onto our model. To computationally quantify glomerular structure despite its complexity, we simplified it to three components consisting of nuclei, capillary lumina and Bowman spaces; and Periodic Acid-Schiff positive structures.We detected glomerular boundaries and nuclei from whole slide images using convolutional neural networks, and the remaining glomerular structures using an unsupervised technique developed expressly for this purpose. We defined a set of digital features which quantify the structural progression of DN, and a recurrent network architecture which processes these features into a classification. Results Our digital classification agreed with a senior pathologist whose classifications were used as ground truth with moderate Cohen's kappa k = 0.55 and 95% confidence interval [0.50, 0.60]. Two other renal pathologists agreed with the digital classification with k150.68, 95% interval [0.50, 0.86] and k2=0.48, 95%interval [0.32, 0.64]. Our results suggest computational approaches are comparable to human visual classificationmethods, and can offer improved precision in clinical decision workflows. We detected glomerular boundaries from whole slide images with 0.9360.04 balanced accuracy, glomerular nuclei with 0.94 sensitivity and 0.93 specificity, and glomerular structural components with 0.95 sensitivity and 0.99 specificity. Conclusions Computationally derived, histologic image features hold significant diagnostic information that may augment clinical diagnostics.

Original languageEnglish (US)
Pages (from-to)1953-1967
Number of pages15
JournalJournal of the American Society of Nephrology
Volume30
Issue number10
DOIs
StatePublished - Jan 1 2019

Fingerprint

Diabetic Nephropathies
Kidney
Sensitivity and Specificity
Periodic Acid
Workflow
Confidence Intervals
Biopsy
Pathologists

ASJC Scopus subject areas

  • Nephrology

Cite this

Computational segmentation and classification of diabetic glomerulosclerosis. / Ginley, Brandon; Lutnick, Brendon; Jen, Kuang Yu; Fogo, Agnes B.; Jain, Sanjay; Rosenberg, Avi; Walavalkar, Vighnesh; Wilding, Gregory; Tomaszewski, John E.; Yacoub, Rabi; Rossi, Giovanni Maria; Sarder, Pinaki.

In: Journal of the American Society of Nephrology, Vol. 30, No. 10, 01.01.2019, p. 1953-1967.

Research output: Contribution to journalArticle

Ginley, B, Lutnick, B, Jen, KY, Fogo, AB, Jain, S, Rosenberg, A, Walavalkar, V, Wilding, G, Tomaszewski, JE, Yacoub, R, Rossi, GM & Sarder, P 2019, 'Computational segmentation and classification of diabetic glomerulosclerosis', Journal of the American Society of Nephrology, vol. 30, no. 10, pp. 1953-1967. https://doi.org/10.1681/ASN.2018121259
Ginley, Brandon ; Lutnick, Brendon ; Jen, Kuang Yu ; Fogo, Agnes B. ; Jain, Sanjay ; Rosenberg, Avi ; Walavalkar, Vighnesh ; Wilding, Gregory ; Tomaszewski, John E. ; Yacoub, Rabi ; Rossi, Giovanni Maria ; Sarder, Pinaki. / Computational segmentation and classification of diabetic glomerulosclerosis. In: Journal of the American Society of Nephrology. 2019 ; Vol. 30, No. 10. pp. 1953-1967.
@article{4e786ee0ae2543438553adaa8bf6a532,
title = "Computational segmentation and classification of diabetic glomerulosclerosis",
abstract = "Background Pathologists use visual classification of glomerular lesions to assess samples from patients with diabetic nephropathy (DN). The results may vary among pathologists. Digital algorithms may reduce this variability and provide more consistent image structure interpretation. Methods We developed a digital pipeline to classify renal biopsies from patients with DN. We combined traditional image analysis with modern machine learning to efficiently capture important structures, minimize manual effort and supervision, and enforce biologic prior information onto our model. To computationally quantify glomerular structure despite its complexity, we simplified it to three components consisting of nuclei, capillary lumina and Bowman spaces; and Periodic Acid-Schiff positive structures.We detected glomerular boundaries and nuclei from whole slide images using convolutional neural networks, and the remaining glomerular structures using an unsupervised technique developed expressly for this purpose. We defined a set of digital features which quantify the structural progression of DN, and a recurrent network architecture which processes these features into a classification. Results Our digital classification agreed with a senior pathologist whose classifications were used as ground truth with moderate Cohen's kappa k = 0.55 and 95{\%} confidence interval [0.50, 0.60]. Two other renal pathologists agreed with the digital classification with k150.68, 95{\%} interval [0.50, 0.86] and k2=0.48, 95{\%}interval [0.32, 0.64]. Our results suggest computational approaches are comparable to human visual classificationmethods, and can offer improved precision in clinical decision workflows. We detected glomerular boundaries from whole slide images with 0.9360.04 balanced accuracy, glomerular nuclei with 0.94 sensitivity and 0.93 specificity, and glomerular structural components with 0.95 sensitivity and 0.99 specificity. Conclusions Computationally derived, histologic image features hold significant diagnostic information that may augment clinical diagnostics.",
author = "Brandon Ginley and Brendon Lutnick and Jen, {Kuang Yu} and Fogo, {Agnes B.} and Sanjay Jain and Avi Rosenberg and Vighnesh Walavalkar and Gregory Wilding and Tomaszewski, {John E.} and Rabi Yacoub and Rossi, {Giovanni Maria} and Pinaki Sarder",
year = "2019",
month = "1",
day = "1",
doi = "10.1681/ASN.2018121259",
language = "English (US)",
volume = "30",
pages = "1953--1967",
journal = "Journal of the American Society of Nephrology : JASN",
issn = "1046-6673",
publisher = "American Society of Nephrology",
number = "10",

}

TY - JOUR

T1 - Computational segmentation and classification of diabetic glomerulosclerosis

AU - Ginley, Brandon

AU - Lutnick, Brendon

AU - Jen, Kuang Yu

AU - Fogo, Agnes B.

AU - Jain, Sanjay

AU - Rosenberg, Avi

AU - Walavalkar, Vighnesh

AU - Wilding, Gregory

AU - Tomaszewski, John E.

AU - Yacoub, Rabi

AU - Rossi, Giovanni Maria

AU - Sarder, Pinaki

PY - 2019/1/1

Y1 - 2019/1/1

N2 - Background Pathologists use visual classification of glomerular lesions to assess samples from patients with diabetic nephropathy (DN). The results may vary among pathologists. Digital algorithms may reduce this variability and provide more consistent image structure interpretation. Methods We developed a digital pipeline to classify renal biopsies from patients with DN. We combined traditional image analysis with modern machine learning to efficiently capture important structures, minimize manual effort and supervision, and enforce biologic prior information onto our model. To computationally quantify glomerular structure despite its complexity, we simplified it to three components consisting of nuclei, capillary lumina and Bowman spaces; and Periodic Acid-Schiff positive structures.We detected glomerular boundaries and nuclei from whole slide images using convolutional neural networks, and the remaining glomerular structures using an unsupervised technique developed expressly for this purpose. We defined a set of digital features which quantify the structural progression of DN, and a recurrent network architecture which processes these features into a classification. Results Our digital classification agreed with a senior pathologist whose classifications were used as ground truth with moderate Cohen's kappa k = 0.55 and 95% confidence interval [0.50, 0.60]. Two other renal pathologists agreed with the digital classification with k150.68, 95% interval [0.50, 0.86] and k2=0.48, 95%interval [0.32, 0.64]. Our results suggest computational approaches are comparable to human visual classificationmethods, and can offer improved precision in clinical decision workflows. We detected glomerular boundaries from whole slide images with 0.9360.04 balanced accuracy, glomerular nuclei with 0.94 sensitivity and 0.93 specificity, and glomerular structural components with 0.95 sensitivity and 0.99 specificity. Conclusions Computationally derived, histologic image features hold significant diagnostic information that may augment clinical diagnostics.

AB - Background Pathologists use visual classification of glomerular lesions to assess samples from patients with diabetic nephropathy (DN). The results may vary among pathologists. Digital algorithms may reduce this variability and provide more consistent image structure interpretation. Methods We developed a digital pipeline to classify renal biopsies from patients with DN. We combined traditional image analysis with modern machine learning to efficiently capture important structures, minimize manual effort and supervision, and enforce biologic prior information onto our model. To computationally quantify glomerular structure despite its complexity, we simplified it to three components consisting of nuclei, capillary lumina and Bowman spaces; and Periodic Acid-Schiff positive structures.We detected glomerular boundaries and nuclei from whole slide images using convolutional neural networks, and the remaining glomerular structures using an unsupervised technique developed expressly for this purpose. We defined a set of digital features which quantify the structural progression of DN, and a recurrent network architecture which processes these features into a classification. Results Our digital classification agreed with a senior pathologist whose classifications were used as ground truth with moderate Cohen's kappa k = 0.55 and 95% confidence interval [0.50, 0.60]. Two other renal pathologists agreed with the digital classification with k150.68, 95% interval [0.50, 0.86] and k2=0.48, 95%interval [0.32, 0.64]. Our results suggest computational approaches are comparable to human visual classificationmethods, and can offer improved precision in clinical decision workflows. We detected glomerular boundaries from whole slide images with 0.9360.04 balanced accuracy, glomerular nuclei with 0.94 sensitivity and 0.93 specificity, and glomerular structural components with 0.95 sensitivity and 0.99 specificity. Conclusions Computationally derived, histologic image features hold significant diagnostic information that may augment clinical diagnostics.

UR - http://www.scopus.com/inward/record.url?scp=85072791378&partnerID=8YFLogxK

UR - http://www.scopus.com/inward/citedby.url?scp=85072791378&partnerID=8YFLogxK

U2 - 10.1681/ASN.2018121259

DO - 10.1681/ASN.2018121259

M3 - Article

C2 - 31488606

AN - SCOPUS:85072791378

VL - 30

SP - 1953

EP - 1967

JO - Journal of the American Society of Nephrology : JASN

JF - Journal of the American Society of Nephrology : JASN

SN - 1046-6673

IS - 10

ER -