Computer-aided Prognosis of Neuroblastoma on Whole-slide Images: Classification of Stromal Development

Abstract

We are developing a computer-aided prognosis system for neuroblastoma (NB), a cancer of the nervous system and one of the most malignant tumors affecting children. Histopathological examination is an important stage for further treatment planning in routine clinical diagnosis of NB. According to the International Neuroblastoma Pathology Classification (the Shimada system), NB patients are classified into favorable and unfavorable histology based on the tissue morphology. In this study, we propose an image analysis system that operates on digitized H&E stained whole-slide NB tissue samples and classifies each slide as either stroma-rich or stroma-poor based on the degree of Schwannian stromal development. Our statistical framework performs the classification based on texture features extracted using co-occurrence statistics and local binary patterns. Due to the high resolution of digitized whole-slide images, we propose a multi-resolution approach that mimics the evaluation of a pathologist such that the image analysis starts from the lowest resolution and switches to higher resolutions when necessary. We employ an offine feature selection step, which determines the most discriminative features at each resolution level during the training step. A modified k-nearest neighbor classifier is used to determine the confidence level of the classification to make the decision at a particular resolution level. The proposed approach was independently tested on 43 whole-slide samples and provided an overall classification accuracy of 88.4%.

Fig. 1

A simplified diagram of the International Neuroblastoma Pathology Classification (the Shimada system), where UH and FH correspond to unfavorable and favorable histology, respectively.

Fig. 2

Example images of (a,b) stroma-rich and (c,d) stroma-poor tissue.

Fig. 3

Computational infrastructure used to process the whole-slide images in parallel.

Fig. 4

Flowchart of the whole-slide image analysis system for NB prognosis. KNN and LBP correspond to k-nearest neighbor and local binary patterns, respectively.

Fig. 5

(a) The conventional LBP operator; (b) circular pattern used to compute rotation invariant uniform patterns

Fig. 6

The 2D scatter plot of feature spaces associated with different resolution levels. The discrimination of the class samples increases proportionally with the increasing resolutions 3 through 0, respectively.

Fig. 7

Flowchart of the training process.

Fig. 8

Sample classification results after processing a whole-slide NB image. (a) and (d) are the H&E stained NB slides associated with stroma-rich and stroma-poor by an expert pathologist. (b) and (e) are the classification maps identified by the computerized system where the red color corresponds to stroma-rich regions and the green color corresponds to stroma-poor regions. (c) and (f) are the corresponding decision level statistics that show in log-scale the number of image tiles classified at a certain resolution level. In the resolution level map on the upper right, cyan color represents the lowest resolution and green color represents the highest resolution, respectively.

Fig. 9

The distribution of the image tiles over the 11 stroma-rich, whole-slide images. The red, green, and white colors represent the number of image tiles classified as stroma-rich, stroma-poor, and background tiles, respectively. The black color represents the number of image tiles that were not able to be classified by the computerized system.

Fig. 10

The distribution of the image tiles over the 32 stroma-poor, whole-slide images. The red, green, and white colors represent the number of image tiles classified as stroma-rich, stroma-poor, and background tiles, respectively. The black color represents the number of image tiles that were not able to be classified by the computerized system.