Student or educator? Sign up with your school email and get 6 months of Pro free.
CoChat

The DOI matches its Crossref record; 49 of the 50 references it cites were found in a registry or an index.

Multi-class texture analysis in colorectal cancer histology

Open access

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.

Overview

The claims and statistics of this paper haven’t been checked yet.

Free: it reads the open-access text and checks the first 5 claims that cite a source we can check. Create a free account to check every claim.

A few free checks a day · papers already verified open instantly · results are public

Integrity signals

Registry

Crossref · 10.1038/srep27988

Cites

49 of 49 references verified0 citation errors · 0 not checkable

Statistics

Figures

Figure 1 Representative images from our dataset. Here, the first 10 images of every tissue class in our dataset are shown. They represent the wide variation of illumination, stain intensity and tissue textures present in routine histopathological images. Images were extracted from 10 independent samples of colorectal cancer (CRC) primary tumours. (a) tumour epithelium, (b) simple stroma, (c) complex stroma (stroma that contains single tumour cells and/or single immune cells), (d) immune cell conglomerates, (e) debris and mucus, (f) mucosal glands, (g) adipose tissue, (h) background.
Figure 1 Representative images from our dataset. Here, the first 10 images of every tissue class in our dataset are shown. They represent the wide variation of illumination, stain intensity and tissue textures present in routine histopathological images. Images were extracted from 10 independent samples of colorectal cancer (CRC) primary tumours. (a) tumour epithelium, (b) simple stroma, (c) complex stroma (stroma that contains single tumour cells and/or single immune cells), (d) immune cell conglomerates, (e) debris and mucus, (f) mucosal glands, (g) adipose tissue, (h) background.CC BY · PubMed Central
Figure 2 Benchmarking of pure and combined feature sets and four classifiers. (a) Experimentally measured error rate in two-category problems (tumour-stroma separation). (b) Error rate in multiclass problems (8 tissue categories). Classification accuracy is given as the mean of a 10 experimental runs with 90% of the dataset as training and 10% of the dataset as testing group (10-fold cross-validation). It can be seen that radial basis function (rbf) support vector machine (SVM) outperforms other classifiers, especially in the multiclass setting. (c) After testing the pure feature sets, we assessed discriminatory power of the concatenated feature vectors. Best2 = histogram-lower and LBP; best3 = best2 and histogram-higher; best4 = best3 and GLCM; best5 = best4 and Perceptual; all6 includes all features. Accuracy reaches an optimum in the best5 set (bar highlighted in yellow). (d) Computational performance is acceptable in best5 set but not in the all6 set. Abbreviations: LBP = local binary patterns, GLCM = gray-level co-occurrence matrix, 1-NN = 1-nearest neighbour, ensTree = ensemble of decision trees, linSVM = linear support vector machine (SVM), rbfSVM = radial basis function SVM.
Figure 2 Benchmarking of pure and combined feature sets and four classifiers. (a) Experimentally measured error rate in two-category problems (tumour-stroma separation). (b) Error rate in multiclass problems (8 tissue categories). Classification accuracy is given as the mean of a 10 experimental runs with 90% of the dataset as training and 10% of the dataset as testing group (10-fold cross-validation). It can be seen that radial basis function (rbf) support vector machine (SVM) outperforms other classifiers, especially in the multiclass setting. (c) After testing the pure feature sets, we assessed discriminatory power of the concatenated feature vectors. Best2 = histogram-lower and LBP; best3 = best2 and histogram-higher; best4 = best3 and GLCM; best5 = best4 and Perceptual; all6 includes all features. Accuracy reaches an optimum in the best5 set (bar highlighted in yellow). (d) Computational performance is acceptable in best5 set but not in the all6 set. Abbreviations: LBP = local binary patterns, GLCM = gray-level co-occurrence matrix, 1-NN = 1-nearest neighbour, ensTree = ensemble of decision trees, linSVM = linear support vector machine (SVM), rbfSVM = radial basis function SVM.CC BY · PubMed Central
Figure 3 Performance of two classification methods for eight tissue categories. (a) Confusion matrix for the best pure feature set histogram-lower, classified with a radial basis function (rbf) support vector machine (SVM), (b) corresponding receiver operating characteristic (ROC) curves (one-against-all approach), mean area under the curve (AUC) = 0.968. (c) Confusion matrix for the best combined feature set (best5); (d) corresponding ROC with mean AUC = 0.976. The confusion matrices show that histogram-lower misclassifies only few samples, but best5 improves this performance even more. Classes: 1 = Tumour epithelium, 2 = Stroma (simple), 3 = Stroma (complex), 4 = Immune cell conglomerates, 5 = Debris and mucus, 6 = Glands, 7 = Adipose, 8 = Background.
Figure 3 Performance of two classification methods for eight tissue categories. (a) Confusion matrix for the best pure feature set histogram-lower, classified with a radial basis function (rbf) support vector machine (SVM), (b) corresponding receiver operating characteristic (ROC) curves (one-against-all approach), mean area under the curve (AUC) = 0.968. (c) Confusion matrix for the best combined feature set (best5); (d) corresponding ROC with mean AUC = 0.976. The confusion matrices show that histogram-lower misclassifies only few samples, but best5 improves this performance even more. Classes: 1 = Tumour epithelium, 2 = Stroma (simple), 3 = Stroma (complex), 4 = Immune cell conglomerates, 5 = Debris and mucus, 6 = Glands, 7 = Adipose, 8 = Background.CC BY · PubMed Central
Figure 4 Application of trained classifiers to unknown images. (a–j) After training and testing classification performance, we applied the trained classifier to a set of unknown images with highly intermixed textures of which this figure shows 5 representative examples. The left panel (a,c,e,g,i) shows the original image and the right panel (b,d,f,h,j) shows the classification result in a colour code (observe the legend in (k)). Each image was divided into 10,000 patches and each patch was assigned to one of eight tissue classes. For this experiment, we used a combined feature set (best5), classifier was a 10-fold cross validated radial basis function (rbf) support vector machine (SVM) with 1-vs-1 class decisions in an error-correcting output code multiclass model (ECOC). Colour-coded classification maps were smoothed by a 3 × 3 median filter and enlarged with bicubic interpolation.
Figure 4 Application of trained classifiers to unknown images. (a–j) After training and testing classification performance, we applied the trained classifier to a set of unknown images with highly intermixed textures of which this figure shows 5 representative examples. The left panel (a,c,e,g,i) shows the original image and the right panel (b,d,f,h,j) shows the classification result in a colour code (observe the legend in (k)). Each image was divided into 10,000 patches and each patch was assigned to one of eight tissue classes. For this experiment, we used a combined feature set (best5), classifier was a 10-fold cross validated radial basis function (rbf) support vector machine (SVM) with 1-vs-1 class decisions in an error-correcting output code multiclass model (ECOC). Colour-coded classification maps were smoothed by a 3 × 3 median filter and enlarged with bicubic interpolation.CC BY · PubMed Central
Figure 5 Class probabilities in an example image. For each of 10,000 image patches, the probability of belonging to any of eight tissue classes is shown for an example image. These probability maps correspond to the “committee votes” that a given classifier casts for each image block. (a–h) Probability maps, (i) original image.
Figure 5 Class probabilities in an example image. For each of 10,000 image patches, the probability of belonging to any of eight tissue classes is shown for an example image. These probability maps correspond to the “committee votes” that a given classifier casts for each image block. (a–h) Probability maps, (i) original image.CC BY · PubMed Central
Figure 6 Alternative visualization of classification results. (a,d,g) Original image, (b,e,h) classification decision, (c,f,i) committee votes. The committee votes for tumour, simple and complex stroma in each image block are shown as a false colour visualization with red = tumour, green = simple stroma, blue = complex stroma. This visualization shows spatial overlap of textures. For some tissue regions, this visualization can clarify classification results better than just showing a single classification decision. For this figure, we used images from the application set that primarily consisted of tumour and stroma. (j) Legend to (b,e,h). (k) Legend to (c,f,i).
Figure 6 Alternative visualization of classification results. (a,d,g) Original image, (b,e,h) classification decision, (c,f,i) committee votes. The committee votes for tumour, simple and complex stroma in each image block are shown as a false colour visualization with red = tumour, green = simple stroma, blue = complex stroma. This visualization shows spatial overlap of textures. For some tissue regions, this visualization can clarify classification results better than just showing a single classification decision. For this figure, we used images from the application set that primarily consisted of tumour and stroma. (j) Legend to (b,e,h). (k) Legend to (c,f,i).CC BY · PubMed Central
Figure 7 Correlation analysis of the final feature vector. (a) Correlation (Pearson’s correlation coefficient) of all 74 features in the best5 feature set. It can be seen that some subsets of best5 are highly internally correlated (histogram-higher and parts of GLCM) but that between the feature subsets, very little correlation is observed. (b) Correlation of all 5000 samples in the image dataset as measures by the correlation of their 74-dimensional feature vector. Already in this very simple non-parametric analysis, the individual tissue groups form visually discernible clusters.
Figure 7 Correlation analysis of the final feature vector. (a) Correlation (Pearson’s correlation coefficient) of all 74 features in the best5 feature set. It can be seen that some subsets of best5 are highly internally correlated (histogram-higher and parts of GLCM) but that between the feature subsets, very little correlation is observed. (b) Correlation of all 5000 samples in the image dataset as measures by the correlation of their 74-dimensional feature vector. Already in this very simple non-parametric analysis, the individual tissue groups form visually discernible clusters.CC BY · PubMed Central

Verification

Claims the paper makes

The claims and statistics of this paper haven’t been checked yet.

Free: it reads the open-access text and checks the first 5 claims that cite a source we can check. Create a free account to check every claim.

A few free checks a day · papers already verified open instantly · results are public

Self-check

Percentages, sums and reported differences re-computed from the paper’s own numbers.

The self-check needs the paper’s full text.

Citations

Cites

Research works printed but not linked: 1

  1. Allwein, E. L., Schapire, R. E. & Singer, Y. Reducing Multiclass to Binary: A Unifying Approach for Margin Classifiers. J Mach Learn Res 1, 113–141 (2000).

Cited by

See all 695

Metadata

Authors
Kather, J. N., Weis, C.-A., Bianconi, F., Melchers, S. M., Schad, L. R., Gaiser, T., Marx, A., & Zöllner, F. G.
Affiliations
  • Heidelberg University (DE)
  • University of Perugia (IT)
Corresponding author
Kather, Jakob Nikolas
Venue
Scientific Reports
Published
2016-06-16
Volume
6
Issue
1
Pages
27988-27988
ISSN
2045-2322 (ISSN-L)
OpenAlex
W2435090885
Cited by
578
References
50
Citations by year
2026: 66 · 2025: 79 · 2024: 86 · 2023: 68 · 2022: 67 · 2021: 72 · 2020: 63 · 2019: 37 · 2018: 21 · 2017: 18
FWCI
20.82
Type
journal article
Language
en
Open access
Licence
Locations
  • Scientific Reports (publishedVersion, cc-by)
  • Europe PMC (PubMed Central) (publishedVersion, cc-by)
  • PubMed Central (publishedVersion, cc-by)
  • PubMed (publishedVersion)
Funders
Oulun Yliopisto
Topics
  • Physical Sciences › Computer Science › Artificial Intelligence › AI in cancer detection
  • Health Sciences › Medicine › Radiology, Nuclear Medicine and Imaging › Radiomics and Machine Learning in Medical Imaging
  • Physical Sciences › Computer Science › Computer Vision and Pattern Recognition › Image Retrieval and Classification Techniques
Keywords
colorectal cancer, histopathology, tissue classification, texture features, digital pathology, public datasets
MeSH terms
Histocytochemistry, Humans, Image Processing, Computer-Assisted, Colorectal Neoplasms, Automation, Laboratory

About these checks

Registry, retraction, Cites and statistics results are computed, not judged, with one exception: a printed reference that no index lists is read once by a model before it counts as a citation error. Claims are extracted by a model and kept only when their quote appears verbatim in the text; each cited source is then checked, with a model’s help, for whether it says what the claim relies on. Those verdicts can be wrong — read the claim’s quote and the cited source before relying on one.

Statistics and arithmetic are recomputed, not judged.

Confidence bands are shown only for fields calibrated on our annotated test set. None is calibrated yet, so no band is shown.

What counts as a claim ›How we measure confidence ›How we check this ›

Are you an author, or do you see a problem with this page? Request a review

Data sources

Bibliographic data: Semantic Scholar, OpenAlex, Crossref.

Contains information from the Semantic Scholar Academic Graph Datasets, made available under the ODC Attribution License (ODC-BY 1.0).

Semantic Scholar’s data is described in: Kinney, R., et al. (2023). The Semantic Scholar Open Data Platform. arXiv:2301.10140.