Stereology and computer-based image analysis quantifies heterogeneity and improves reproducibility for grading reticulin in myeloproliferative neoplasms

Abstract

Evaluation of bone marrow fibrosis in myeloproliferative neoplasms (MPN) is subject to interobserver inconsistency, and cutpoints for determination of therapy are dependent on manual pathologist grading of reticulin. To provide a more consistent approach to cutpoint determination, we developed a stereology-based method of calculating length fiber density (the length of the fiber network divided by the volume of bone marrow hematopoietic tissue). Forty-eight thin needle core bone marrow biopsy samples from patients with MPN were obtained from the University of Utah archives after institutional review board approval and stained for reticulin and then scanned using whole slide imaging. To determine interpathologist concordance, the blinded cases were scored according to the European consensus system by four pathologists. Two pathologists scored twice with a 1-month washout period to determine intrapathologist concordance. Using systematic uniform random sampling and line counting, two techniques adapted from the stereology field, the length density of the reticulin network was measured as well as a measure of heterogeneity across the bone marrow sample. Image analysis was used to measure the area of the reticulin fiber as a two-dimensional profile. To build a model for predicting scoring by stereology, 16 samples were chosen randomly and a linear regression relationship determined. This score was then rounded to 0, 1, 2, or 3, and concordance rates between pathologists and between pathologist and the computer stereology score determined.: In the 48 patient cohort, stereology assessment was well correlated with the average of manual pathologist scoring (linear regression, 2=0.7038). While the European consensus scoring system ranges from 0 to 3, the cutpoint between 0/1 and 2/3 is generally used for guiding therapeutic decision as myelofibrosis. In 7 of the 48 cases (15 %), pathologists had differing scores across this cutpoint, meaning that the ultimate therapeutic outcome would have been impacted. The stereology score differed with the average of the pathologist scores in only 5 of the 48 cases across this cutpoint (10 %). Precision rates for the stereology analysis were 92 %, compared with interpathologist concordance ranging from 54.2 to 65.2 % and intrapathologist at 60.9 %. Computer-based stereology proved to be more reproducible at predicting therapeutic cutpoint than manual scoring. The new technique can be run using standard histochemistry and provides both a nonbiased systematic measure of reticulin and a new measure of reticulin heterogeneity.