Interobserver variability between experienced and inexperienced observers in the histopathological analysis of Wilms tumors: a pilot study for future algorithmic approach

Jikke J Rutgers,Annelies M C Mavinkurve-Groothuis,Marijn A Scheijde-Vermeulen,Ananda Van Der Kamp,Marta Fiocco,Ronald R De Krijger,Tessa Bánki,Tomas J Waterlander,Jeroen A W M Van Der Laak,Marry M Van Den Heuvel-Eibrink

doi:10.1186/s13000-021-01136-w

Abstract

BackgroundHistopathological classification of Wilms tumors determines treatment regimen. Machine learning has been shown to contribute to histopathological classification in various malignancies but requires large numbers of manually annotated images and thus specific pathological knowledge. This study aimed to assess whether trained, inexperienced observers could contribute to reliable annotation of Wilms tumor components for classification performed by machine learning.MethodsFour inexperienced observers (medical students) were trained in histopathology of normal kidneys and Wilms tumors by an experienced observer (pediatric pathologist). Twenty randomly selected scanned Wilms tumor-slides (from n = 1472 slides) were annotated, and annotations were independently classified by both the inexperienced observers and two experienced pediatric pathologists. Agreement between the six observers and for each tissue element was measured using kappa statistics (κ).ResultsPairwise interobserver agreement between all inexperienced and experienced observers was high (range: 0.845–0.950). The interobserver variability for the different histological elements, including all vital tumor components and therapy-related effects, showed high values for all κ-coefficients (> 0.827).ConclusionsInexperienced observers can be trained to recognize specific histopathological tumor and tissue elements with high interobserver agreement with experienced observers. Nevertheless, supervision by experienced pathologists remains necessary. Results of this study can be used to facilitate more rapid progress for supervised machine learning-based algorithm development in pediatric pathology and beyond.

Highlights

Wilms tumors (WTs) account for approximately 90% of all pediatric renal tumors
Histological classification is the cornerstone of current risk stratification in WTs, thereby defining individual treatment regimens [5]
Study population Twenty whole slide images (WSIs) were randomly selected from the total scanned series of 1472 WSIs. These 20 slides were derived from the resection specimens of 20 WT patients with a total of 22 tumors classified according to the Society for Pediatric Oncology (SIOP)-2001 and the Umbrella SIOP-RTSG 2016 histological classification [5]

Summary

Introduction

Wilms tumors (WTs) account for approximately 90% of all pediatric renal tumors. Since renal tumors account for only 6% of all pediatric malignancies, these tumors are rare [1]. There are significant differences in survival within this group of patients, indicating the need for risk stratification. Histological classification is the cornerstone of current risk stratification in WTs, thereby defining individual treatment regimens [5]. Histopathological classification of Wilms tumors determines treatment regimen. Machine learning has been shown to contribute to histopathological classification in various malignancies but requires large numbers of manually annotated images and specific pathological knowledge. This study aimed to assess whether trained, inexperienced observers could contribute to reliable annotation of Wilms tumor components for classification performed by machine learning

Objectives

Methods

Results

Discussion

Conclusion