Pathological prognosis classification of patients with neuroblastoma using computational pathology analysis

Highlights

• INPC is a time-consuming and laborious process in pathology diagnosis.

• Computational pathology plays a crucial role in better neuroblastoma quantification.

• This study predicted the pathological prognosis using machine learning methods.

• Nucleus intensity features could predict the pathological prognosis in neuroblastoma.

Abstract

Neuroblastoma is the most common extracranial solid tumor in early childhood. International Neuroblastoma Pathology Classification (INPC) is a commonly used classification system that provides clinicians with a reference for treatment stratification. However, given the complex and subjective assessment of the INPC, there will be inconsistencies in the analysis of the same patient by multiple pathologists. An automated, comprehensive and objective classification method is needed to identify different prognostic groups in patients with neuroblastoma. In this study, we collected 563 hematoxylin and eosin-stained histopathology whole-slide images from 107 patients with neuroblastoma who underwent surgical resection. We proposed a novel processing pipeline for nuclear segmentation, cell-level image feature extraction, and patient-level feature aggregation. Logistic regression model was built to classify patients with favorable histology (FH) and patients with unfavorable histology (UH). On the training/test dataset, patient-level of nucleus morphological/intensity features and age could correctly classify patients with a mean area under the receiver operating characteristic curve (AUC) of 0.946, a mean accuracy of 0.856, and a mean Matthews Correlation Coefficient (MCC) of 0.703,respectively. On the independent validation dataset, the classification model achieved a mean AUC of 0.938, a mean accuracy of 0.865 and a mean MCC of 0.630, showing good generalizability. Our results suggested that automatically derived image features could identify the differences in nuclear morphological and intensity between different prognostic groups, which could provide a reference to pathologists and facilitate the evaluation of the pathological prognosis in patients with neuroblastoma.

Introduction

Neuroblastoma, the most common solid extracranial neoplasm in children, accounts for ∼11% of all pediatric cancers but 15% of all pediatric cancer deaths [1,2]. Approaches to risk classification and treatment stratification for neuroblastoma children have led to reduced therapy for low- and intermediate-risk patients and improved outcome for all patients [3,4]. International Neuroblastoma Pathology Classification (INPC) is one of the most powerful prognostic factors in patients with neuroblastoma and can provide clinicians with reference to treatment stratification [5,6].

INPC distinguishes a favorable histology (FH) and an unfavorable histology (UH) using factors such as age, grade of differentiation, mitosis-karyorrhexis index (MKI), and histologic category acquired from whole-slide images (WSIs) [5]. While grade is a standardized way of measuring differentiation, MKI is defined as the number of mitotic tumor cells in the process of karyorrhexis among 5000 tumor cells [7], and histological category refers to the classification of neuroblastoma into different subtypes according to percentage of cells with ganglion cell-like differentiation [7]. Schwannian stroma is composed of immature Schwann cells. The cells are long spindle-shaped, with rich cytoplasm and light staining, and fine nuclear chromatin. Undifferentiation, poor Schwannian stroma, high MKI, and elder diagnosis age (>5 years old) are associated with lower event-free survival [5]. Due to the characteristics of clinical heterogeneity of neuroblastoma, the degree of differentiation and MKI value at different tumor sites exhibit markedly different properties [8]. Pathologists need to evaluate the cell morphology and cell distribution in multiple WSIs to make an accurate diagnosis. Additionally, MKI involves a manual count of sufficient high-power fields to include 5000 cells, which makes the diagnostic process time-consuming [9]. Therefore, there is an urgent need for an automated, comprehensive and objective prognostic grouping assistive tool in neuroblastoma.

Computational pathology plays a crucial role in better quantification of disease and precision medicine, which can make the manual, subjective and laborious process in pathology diagnosis become more automated, standardized and highly efficient [10]. Computational pathology has been used to perform detailed spatial analysis of the entire tumor microenvironment composition and cells (such as capturing nuclear shape, structure and distribution) [11]. Recently, researchers used computational pathology and machine learning methods for the diagnosis, grading, prognosis, and prediction of the response to cancer therapy, including prostate, breast, oropharyngeal carcinomas, and brain tumors [[12], [13], [14]]. The histological examination of neuroblastoma in children requires comprehensive observation and estimation of tumor components across multiple WSIs, making computational pathology an effective auxiliary tool for predicting the prognosis of neuroblastoma patients. However, few studies addressed this issue.

In this study, we used machine learning techniques and computational pathology to predict the pathological prognosis of neuroblastoma patients based on quantitative image features extracted from segmented nuclei of multiple WSIs. Specifically, we first evaluated the selection of parameters in model performance to obtain the prognostic classification model. Then, risk heatmaps, which assessed the relative importance of different regions of WSIs for prognosis classification were generated.