Multi-class texture analysis in colorectal cancer histology

Abstract

Automatic recognition of different tissue types in histological images is an essential part in the digital pathology toolbox. Texture analysis is commonly used to address this problem; mainly in the context of estimating the tumour/stroma ratio on histological samples. However, although histological images typically contain more than two tissue types, only few studies have addressed the multi-class problem. For colorectal cancer, one of the most prevalent tumour types, there are in fact no published results on multiclass texture separation. In this paper we present a new dataset of 5,000 histological images of human colorectal cancer including eight different types of tissue. We used this set to assess the classification performance of a wide range of texture descriptors and classifiers. As a result, we found an optimal classification strategy that markedly outperformed traditional methods, improving the state of the art for tumour-stroma separation from 96.9% to 98.6% accuracy and setting a new standard for multiclass tissue separation (87.4% accuracy for eight classes). We make our dataset of histological images publicly available under a Creative Commons license and encourage other researchers to use it as a benchmark for their studies.

Computer-aided classification of sickle cell retinopathy using quantitative features in optical coherence tomography angiography

“…Therefore, we speculated that the classification performance could be improved by comprehensive involvement of multiple features. Combined feature analysis have been used in colorectal cancer histology [38]. We used each of the single parameters and trained an SVM classifier to identify control and two stages of SCR.…”

“…For both types of classification, we quantitatively compared the performance of three classifiers. Discriminant analysis and KNN are two most common and basic machine learning algorithms [30,31,38,39]. We wanted to compare these basic algorithms to SVM which is known to be robust for small data set, and has been a widely used algorithm in other research fields [31,39].…”

“…In mild vs. severe classification, with lower data points, even a few outliers in feature vectors resulted in lower performance of KNN. In contrary, SVM works well even in case of nonlinear data points and small image data sets as it demonstrates less model assumption, less outlier sensitivity and less nearby point dependency [31,34,38,40]. Along with our cross validation technique SVM limits the tendency of overfitting data more effectively compared to KNN and discriminant analysis [34,40].…”

Dictionary Learning and Sparse Representation for Image Analysis with Application to Segmentation, Classification and Event Detection

“…Moreover, image intensity or color histograms are not sufficient to discriminate between images, specially in case of Hematoxylin and Eosin (H&E) histo-pathological tissue images. In [157] it was demonstrated that texture features can efficiently discriminate between different tissue types. Also in case of differentiating between military camouflage uniform patterns or discriminating between military and civil vehicles, texture patterns play a key role.…”

“…For example, in differentiation between malignant and benign tumor, often the smoothness of the nuclei are taken into account [159]. While in other applications there different types to tissues are required to be classified [160] based on their textural patterns to determine cellular composition.For example, in Fig. 5.3 (e) and (f), the samples from a colorectal tissue image dataset, the saliency maps make the structural characteristics of the images more prominent.…”