Comparing LBP, HOG and Deep Features for Classification of Histopathology Images

Alhindi, Taha J.; Kalra, Shivam; Ng, Ka Hin; Afrin, Anika; Tizhoosh, Hamid R.

doi:10.1109/ijcnn.2018.8489329

Cited by 79 publications

(50 citation statements)

References 25 publications

(29 reference statements)

Supporting

Mentioning

Contrasting

Unclassified

Order By: Relevance

“…We achieved very good classification performance, with AUCs between 0.99 and 1 for distinguishing normal and tumor samples. Comparing this performance to previous work, we note that in one study of histopathology images [42], classification performance reached 81.14% accuracy using the extracted features from a pre-trained VGG 19 (similar to VGG 16) network. In a similar study of histopathological images of breast cancer [43], classification performance on 400 HE-stained images of 2048 1536 pixels each reached an AUC of 0.963 for distinguishing between non-carcinomas vs. carcinomas samples.…”

Section: Discussionsupporting

confidence: 50%

Prognostic Analysis of Histopathological Images Using Pre-Trained Convolutional Neural Networks

Daigle

2019

Preprint

View full text Add to dashboard Cite

Background: Histopathological images contain rich phenotypic descriptions of the molecular processes underlying disease progression. Convolutional neural networks (CNNs), a state-of-the-art image analysis technique in computer vision, automatically learns representative features from such images which can be useful for disease diagnosis, prognosis, and subtyping. Despite hepatocellular carcinoma (HCC) being the sixth most common type of primary liver malignancy with a high mortality rate, little previous work has made use of CNN models to delineate the importance of histopathological images in diagnosis and clinical survival of HCC. Results:We applied three pre-trained CNN models -VGG 16, Inception V3, and ResNet 50 -to extract features from HCC histopathological images. The visualization and classification showed clear separation between cancer and normal samples using image features. In a univariate Cox regression analysis, 21.4% and 16% of image features on average were significantly associated with overall survival and disease-free survival, respectively. We also observed significant correlations between these features and integrated biological pathways derived from gene expression and copy number variation. Using an elastic net regularized CoxPH model of overall survival, we obtained a concordance index (C-index) of 0.789 and a significant log-rank test (p = 7.6E-18) after applying Inception image features. We also performed unsupervised classification to identify HCC subgroups from image features. The optimal two subgroups discovered using Inception image features were significantly associated with both overall (C-index = 0.628 and p = 7.39E-07) and disease-free survival (C-index = 0.558 and p = 0.012). Our results suggest the feasibility of feature extraction using pre-trained models, as well as the utility of the resulting features to build an accurate prognosis model of HCC and highlight significant correlations with clinical survival and biological pathways. Conclusions:The image features extracted from HCC histopathological images using the pre-trained CNN models VGG 16, Inception V3 and ResNet 50 can accurately distinguish normal and cancer samples. Furthermore, these image features are significantly correlated with relevant biological outcomes.

show abstract

Section: Discussionsupporting

confidence: 50%

Prognostic Analysis of Histopathological Images Using Pre-Trained Convolutional Neural Networks

Daigle

2019

Preprint

View full text Add to dashboard Cite

show abstract

“…To intuitively compare the effects of our feature with other mainstream features (LBP [14], HOG [13], GLCM [15], FHOG [29]) in the task of floating object detection, a comparative experiment is conducted. The experiment occurred under seven challenging water scenes, including large waves, circular ripples, spray, turbulent water surface, strip ripples, near-view reflection, and far-view reflection.…”

Section: Intuitive Results Of Texture Detectionmentioning

confidence: 99%

“…Traditional methods of combining hand-crafted features with classifiers are simple and fast [12]. Hand-crafted texture features with grayscale invariance, e.g., Histogram of Oriented Gradient (HOG) [13], Local Binary Pattern (LBP) [14], and Gray Level Co-occurrence Matrix (GLCM) [15], are suitable for describing floating objects under uneven illumination. But these features do not utilize global information, making feature descriptors inaccurate when other sever interferences exist.…”

mentioning

confidence: 99%

Combination of Spatial and Frequency Domains for Floating Object Detection on Complex Water Surfaces

et al. 2019

View full text Add to dashboard Cite

Featured Application: We proposed a method of floating object detection, which is characterized by small sample training, strong anti-interference ability, low time consumption and high precision. The method combines spatial-based texture detection and frequency-based saliency detection. Our research is based on the natural images collected by an unmanned surface cleaning robot, which is one of the important applications in water source conservation. Abstract: In order to address the problems of various interference factors and small sample acquisition in surface floating object detection, an object detection algorithm combining spatial and frequency domains is proposed. Firstly, a rough texture detection is performed in a spatial domain. A Fused Histogram of Oriented Gradient (FHOG) is combined with a Gray Level Co-occurrence Matrix (GLCM) to describe global and local information of floating objects, and sliding windows are classified by Support Vector Machines (SVM) with new texture features. Then, a novel frequency-based saliency detection method used in complex scenes is proposed. It adopts global and local low-rank decompositions to remove redundant regions caused by multiple interferences and retain floating objects. The final detection result is obtained by a strategy of combining bounding boxes from different processing domains. Experimental results show that the overall performance of the proposed method is superior to other popular methods, including traditional image segmentation, saliency detection, hand-crafted texture detection, and Convolutional Neural Network Based (CNN-based) object detection. The proposed method is characterized by small sample training and strong anti-interference ability in complex water scenes like ripple, reflection, and uneven illumination. The average precision of the proposed is 97.2%, with only 0.504 seconds of time consumption.Few researchers have focused on floating object detection with respect to complex water surfaces. In general, the existing methods that can be applied to the detection task fall into four categories: traditional image segmentation, saliency detection, hand-crafted feature detection, and object detection based on a Convolutional Neural Network (CNN). According to different processing domains, these methods can also be classified into spatial-based methods and frequency-based methods.In spatial domain, traditional image segmentation has an excellent real-time performance [4]. Wang et al.[5] extracted contours of floating objects according to image grayscale, then framed the contours that met a special size criterion. Xue et al. [6] used the Super-pixel Merging method to segment eligible foreground regions. Tang et al. [7] used Mean-shift clustering and an improved Otsu [8] method to detect floating objects. Although these methods work well in calm water, they are not robust enough when ripples or reflections exist. To this end, Jin et al. [9] proposed an improved Gaussian Mixture Model Based (GMM-based) automatic segmentation method (IGASM) to dete...

show abstract

“…Also, Kiefer et al[31] explored the use of deep features from several 153 pre-trained structures on Kimia24, controlling for the impact of transfer learning and finding an advantage of pre-trained networks against training from scratch. Alhindi and 155 colleagues[32] instead analyze Kimia960 for slide of origin (20 slides preselected by 156 visual ispection), and similarly to our study compare alternative classifiers as well as 157 feature extraction models in a 3-fold CV setup. Our framework differs for the DAP 158 structure and is originally applied to the larger HINT set without any visual 159…”

mentioning

confidence: 76%

“…Deep learning refers to a class of machine 28 learning methods that model high-level abstractions in data through the use of modular 29 architectures, typically composed by multiple nonlinear transformations estimated by 30 training procedures. Notably, deep learning architectures based on Convolutional 31 Neural Networks (CNNs) hold state-of-the-art accuracy in numerous image classification 32 tasks without prior feature selection. Further, intermediate steps in the pipeline of 33 transformations implemented by CNNs or other deep learning architectures can provide 34 a mapping (embedding) from the original feature space into a deep feature space.…”

mentioning

confidence: 99%

Evaluating reproducibility of AI algorithms in digital pathology with DAPPER

Bizzego

Bussola

Chierici

et al. 2018

Preprint

View full text Add to dashboard Cite

Artificial Intelligence is exponentially increasing its impact on healthcare. As deep learning is mastering computer vision tasks, its application to digital pathology is natural, with the promise of aiding in routine reporting and standardizing results across trials. Deep learning features inferred from digital pathology scans can improve validity and robustness of current clinico-pathological features, up to identifying novel histological patterns, e.g. from tumor infiltrating lymphocytes. In this study, we examine the issue of evaluating accuracy of predictive models from deep learning features in digital pathology, as an hallmark of reproducibility. We introduce the DAPPER framework for validation based on a rigorous Data Analysis Plan derived from the FDA's MAQC project, designed to analyse causes of variability in predictive biomarkers. We apply the framework on models that identify tissue of origin on 787 Whole Slide Images from the Genotype-Tissue Expression (GTEx) project. We test 3 different deep learning architectures (VGG, ResNet, Inception) as feature extractors and three classifiers (a fully connected multilayer, Support Vector Machine and Random Forests) and work with 4 datasets (5, 10, 20 or 30 classes), for a total 53000 tiles at 512 × 512 resolution. We analyze accuracy and feature stability of the machine learning classifiers, also demonstrating the need for random features and random labels diagnostic tests to identify selection bias and risks for reproducibility. Further, we use the deep features from the VGG model from GTEx on the KIMIA24 dataset for identification of slide of origin (24 classes) to train a classifier on 1060 annotated tiles and validated on 265 unseen ones. The DAPPER software, including its deep learning backbone pipeline and the HINT (Histological Imaging -Newsy Tiles) benchmark dataset derived from GTEx, is released as a basis for standardization and validation initiatives in AI for Digital Pathology. Author summaryIn this study, we examine the issue of evaluating accuracy of predictive models from

show abstract

Comparing LBP, HOG and Deep Features for Classification of Histopathology Images

Cited by 79 publications

References 25 publications

Prognostic Analysis of Histopathological Images Using Pre-Trained Convolutional Neural Networks

Prognostic Analysis of Histopathological Images Using Pre-Trained Convolutional Neural Networks

Combination of Spatial and Frequency Domains for Floating Object Detection on Complex Water Surfaces

Evaluating reproducibility of AI algorithms in digital pathology with DAPPER

Contact Info

Product

Resources

About