Application of Convolutional Neural Network  for Histopathological Analysis

Hlavcheva, Daria; Yaloveha, Vladyslav; Podorozhniak, Andrii

doi:10.20998/2522-9052.2019.4.10

Cited by 5 publications

(6 citation statements)

References 17 publications

(19 reference statements)

Supporting

Mentioning

Contrasting

Order By: Relevance

“…We further tested whether purity was being predicted from only the image regions containing individual nuclei, or whether intercellular information was being used. For this, we made use of a CNN classifier 40 that predicts tumor/normal status from individual nucleus images (see "Methods"). We trained on the breast nuclei, and this was able to predict tumor status of reserved breast nuclei images with high accuracy (AUC 93-98%) However, the breast-trained CNN yielded poor predictions on the colorectal nuclei (AUC 56%).…”

Section: K I C H K I R C K I R P L I H C L U a D L U S C O V P A A mentioning

confidence: 99%

See 1 more Smart Citation

Deep learning-based cross-classifications reveal conserved spatial behaviors within tumor histological images

et al. 2020

View full text Add to dashboard Cite

Histopathological images are a rich but incompletely explored data type for studying cancer. Manual inspection is time consuming, making it challenging to use for image data mining. Here we show that convolutional neural networks (CNNs) can be systematically applied across cancer types, enabling comparisons to reveal shared spatial behaviors. We develop CNN architectures to analyze 27,815 hematoxylin and eosin scanned images from The Cancer Genome Atlas for tumor/normal, cancer subtype, and mutation classification. Our CNNs are able to classify TCGA pathologist-annotated tumor/normal status of whole slide images (WSIs) in 19 cancer types with consistently high AUCs (0.995 ± 0.008), as well as subtypes with lower but significant accuracy (AUC 0.87 ± 0.1). Remarkably, tumor/normal CNNs trained on one tissue are effective in others (AUC 0.88 ± 0.11), with classifier relationships also recapitulating known adenocarcinoma, carcinoma, and developmental biology. Moreover, classifier comparisons reveal intra-slide spatial similarities, with an average tile-level correlation of 0.45 ± 0.16 between classifier pairs. Breast cancers, bladder cancers, and uterine cancers have spatial patterns that are particularly easy to detect, suggesting these cancers can be canonical types for image analysis. Patterns for TP53 mutations can also be detected, with WSI self- and cross-tissue AUCs ranging from 0.65-0.80. Finally, we comparatively evaluate CNNs on 170 breast and colon cancer images with pathologist-annotated nuclei, finding that both cellular and intercellular regions contribute to CNN accuracy. These results demonstrate the power of CNNs not only for histopathological classification, but also for cross-comparisons to reveal conserved spatial behaviors across tumors.

show abstract

Section: K I C H K I R C K I R P L I H C L U a D L U S C O V P A A mentioning

confidence: 99%

“…Nucleus-based purity estimation. We implemented the network of Hlavcheva et al 40 , including their reported hyperparameters. The goal of this method is to classify individual nuclei as tumor or normal.…”

Section: Mutational Classificationmentioning

confidence: 99%

Deep learning-based cross-classifications reveal conserved spatial behaviors within tumor histological images

et al. 2020

View full text Add to dashboard Cite

show abstract

“…We tested whether purity estimates were due only to local information around each cell nucleus or whether other image properties were informative. For this we first evaluated a CNN classifier (Hlavcheva et al 2019) designed to predict tumor/normal status from individual nucleus images (see Methods). We trained on the breast nuclei, and this yielded high accuracy on reserved breast nuclei images (AUC 97-99%) However, the breast-trained CNN yielded poor classifications on the colorectal nuclei (AUC 56%), a much worse cross-classification than tile-based WSI-level analysis of TCGA data (Figure 4).…”

Section: Features Impacting Tumor Purity Predictionmentioning

confidence: 99%

“…We implemented the network of Hlavcheva et al (2019) including their reported hyperparameters. The goal of this method is to classify individual nuclei as tumor or normal.…”

Section: Nucleus-based Purity Estimationmentioning

confidence: 99%

Deep learning-based cross-classifications reveal conserved spatial behaviors within tumor histological images

Noorbakhsh

Farahmand

pour

et al. 2019

Preprint

View full text Add to dashboard Cite

Histopathological images are a rich but incompletely explored data type for studying cancer. Manual inspection is time consuming, making it challenging to use for image data mining. Here we show that convolutional neural networks (CNNs) can be systematically applied across cancer types, enabling comparisons to reveal shared spatial behaviors. We develop CNN architectures to analyze 27,815 hematoxylin and eosin slides from The Cancer Genome Atlas for tumor/normal, cancer subtype, and mutation classification. Our CNNs are able to classify tumor/normal status of whole slide images (WSIs) in 19 cancer types with consistently high AUCs (0.995±0.008), as well as subtypes with lower but significant accuracy (AUC 0.87±0.1). Remarkably, tumor/normal CNNs trained on one tissue are effective in others (AUC 0.88±0.11), with classifier relationships also recapitulating known adenocarcinoma, carcinoma, and developmental biology. Moreover, classifier comparisons reveal intra-slide spatial similarities, with average tile-level correlation of 0.45±0.16 between classifier pairs. Breast cancers, bladder cancers, and uterine cancers have spatial patterns that are particularly easy to detect, suggesting these cancers can be canonical types for image analysis. Patterns for TP53 mutations can also be detected, with WSI self-and cross-tissue AUCs ranging from 0.65-0.80. Finally, we comparatively evaluate CNNs on 170 breast and colon cancer images with pathologist-annotated nuclei, finding that both cellular and intercellular regions contribute to CNN accuracy. These results demonstrate the power of CNNs not only for histopathological classification, but also for cross-comparisons to reveal conserved spatial biology.

show abstract

“…According to the previous authors' work [28] resulted classification accuracy on test data has been increased from 0.935 to 0.972. It can be explained by these reasons:…”

Section: Resultsmentioning

confidence: 94%

A Comparison of Classifiers Applied to the Problem of Biopsy Images Analysis

Hlavcheva

Yaloveha

Podorozhniak

et al. 2020

A.I.S.

Self Cite

View full text Add to dashboard Cite

Ab s t r a c t. The purpose of the research is to compare classification algorithms for the histopathological images analyzing issue and to optimize the parameters for obtaining better classification accuracy. The following tasks are solved in the article: preprocessing of BreCaHAD dataset images, implementation and training of CNN, applying K-nearest neighbours, SVM, Random Forest, XGBoost, and perceptron algorithms for classifying features that were extracted by CNN, and results comparison. The object of the research is the process of classifying tumor cells in the microscopic biopsy images. The subject of the research is the process of using ML algorithms for classification of the features extracted by CNN from input biopsy image. The scientific novelty of the research is a comparative analysis of classifiers on the task of "tumor" and "healthy" cells images classification from processed BreCaHAD dataset. As a result it was obtained that from chosen classifiers SVM reached the highest accuracy on test data -0.972. This is the only algorithm that shows better accuracy than perceptron. Perceptron gets 0.966 classification accuracy. K-nearest neighbours, Random Forest, and XGBoost algorithms reached lower results. The algorithms' hyperparameters optimization was carried out. The results have been compared with related works. The following research methods are used: the theory of deep learning, mathematical statistics, parameters optimization.

show abstract

Application of Convolutional Neural Network for Histopathological Analysis

Cited by 5 publications

References 17 publications

Deep learning-based cross-classifications reveal conserved spatial behaviors within tumor histological images

Deep learning-based cross-classifications reveal conserved spatial behaviors within tumor histological images

Deep learning-based cross-classifications reveal conserved spatial behaviors within tumor histological images

A Comparison of Classifiers Applied to the Problem of Biopsy Images Analysis

Contact Info

Product

Resources

About