Showing papers by "Qi Zhang published in 2017"

Histopathological image classification with bilinear convolutional neural networks

Abstract: The computer-aided quantitative analysis for histopathological images has attracted considerable attention. The stain decomposition on histopathological images is usually recommended to address the issue of co-localization or aliasing of tissue substances. Although the convolutional neural networks (CNN) is a popular deep learning algorithm for various tasks on histopathological image analysis, it is only directly performed on histopathological images without considering stain decomposition. The bilinear CNN (BCNN) is a new CNN model for fine-grained classification. BCNN consists of two CNNs, whose convolutional-layer outputs are multiplied with outer product at each spatial location. In this work, we propose a novel BCNN-based method for classification of histopathological images, which first decomposes histopathological images into hematoxylin and eosin stain components, and then perform BCNN on the decomposed images to fuse and improve the feature representation performance. The experimental results on the colorectal cancer histopathological image dataset with eight classes indicate that the proposed BCNN-based algorithm is superior to the traditional CNN.

Sonoelastomics for Breast Tumor Classification: A Radiomics Approach with Clustering-Based Feature Selection on Sonoelastography

Abstract: A radiomics approach to sonoelastography, called "sonoelastomics," is proposed for classification of benign and malignant breast tumors. From sonoelastograms of breast tumors, a high-throughput 364-dimensional feature set was calculated consisting of shape features, intensity statistics, gray-level co-occurrence matrix texture features and contourlet texture features, which quantified the shape, hardness and hardness heterogeneity of a tumor. The high-throughput features were then selected for feature reduction using hierarchical clustering and three-feature selection metrics. For a data set containing 42 malignant and 75 benign tumors from 117 patients, seven selected sonoelastomic features achieved an area under the receiver operating characteristic curve of 0.917, an accuracy of 88.0%, a sensitivity of 85.7% and a specificity of 89.3% in a validation set via the leave-one-out cross-validation, revealing superiority over the principal component analysis, deep polynomial networks and manually selected features. The sonoelastomic features are valuable in breast tumor differentiation.

Histopathological Image Classification With Color Pattern Random Binary Hashing-Based PCANet and Matrix-Form Classifier

Abstract: The computer-aided diagnosis for histopathological images has attracted considerable attention. Principal component analysis network (PCANet) is a novel deep learning algorithm for feature learning with the simple network architecture and parameters. In this study, a color pattern random binary hashing-based PCANet (C-RBH-PCANet) algorithm is proposed to learn an effective feature representation from color histopathological images. The color norm pattern and angular pattern are extracted from the principal component images of R, G, and B color channels after cascaded PCA networks. The random binary encoding is then performed on both color norm pattern images and angular pattern images to generate multiple binary images. Moreover, we rearrange the pooled local histogram features by spatial pyramid pooling to a matrix-form for reducing the dimension of feature and preserving spatial information. Therefore, a C-RBH-PCANet and matrix-form classifier-based feature learning and classification framework is proposed for diagnosis of color histopathological images. The experimental results on three color histopathological image datasets show that the proposed C-RBH-PCANet algorithm is superior to the original PCANet and other conventional unsupervised deep learning algorithms, while the best performance is achieved by the proposed feature learning and classification framework that combines C-RBH-PCANet and matrix-form classifier.

CEUS-based classification of liver tumors with deep canonical correlation analysis and multi-kernel learning

Abstract: The contrast-enhanced ultrasound (CEUS) has been a widely accepted imaging modality for diagnosis of liver cancers. In clinical practice, several typical images selected from enhancement patterns of the arterial, portal venous and late phases can provide reliable information basis for diagnosis. In this work, we propose to develop a CEUS-based computer-aided diagnosis (CAD) for liver cancers with only three typical CEUS images selected from three phases, which simulates the clinical diagnosis mode of radiologists. In the proposed CAD, the deep canonical correlation analysis (DCCA) is first performed on three CEUS pairs between arterial and portal venous phases, arterial and late phases, respectively, due to the effectiveness of multi-view fusion of DCCA. The generated six-view features are then fed to a multiple kernel learning (MKL) classifier to further promote the predictive diagnosis result. The experimental results indicate that the proposed DCCA-MKL algorithm achieves best performance for discriminating benign liver tumors from malignant liver cancers.

Improving MRI-based diagnosis of Alzheimer's disease via an ensemble privileged information learning algorithm

Abstract: In clinical practice, the magnetic resonance imaging (MRI) is a prevalent neuroimaging technique for Alzheimer's disease (AD) diagnosis. As a learning using privileged information (LUPI) algorithm, SVM+ has shown its effectiveness on the classification of brain disorders, with single-modal neuroimaging samples for testing but multimodal neuroimaging samples for training. In this work, we propose to apply the multimodal restricted Boltzmann machines (RBM) as an LUPI algorithm for feature learning so as to form an RBM+ algorithm. Furthermore, an ensemble LUPI algorithm is developed, integrating SVM+ and RBM+ by the multiple kernel boosting based strategy. The experimental results demonstrate that the proposed RBM+ works well as an LUPI algorithm for feature learning, and the ensemble LUPI algorithm is superior to the traditional predictive models for the MRI-based AD diagnosis using the positron emission tomography as the privileged information.

Sonoelastography shows that Achilles tendons with insertional tendinopathy are harder than asymptomatic tendons

Abstract: Purpose To seek differences of Achilles tendon hardness between insertional tendinopathy (IT) and asymptomatic controls by using computer-assisted quantification on axial-strain sonoelastography (ASE).

[Evaluation of Axillary Lymph Node Metastasis by Using Radiomics of Dual-modal Ultrasound Composed of Elastography and B-mode].

Abstract: Objectives To explore the diagnostic value of quantitative radiomics features from dual-modal ultrasound composed of elastography and B-mode for axillary lymph node metastasis in breast cancer patients. Methods We retrospectively analyzed 161 axillary lymph nodes (69 benign and 92 metastatic) undergoing real-time elastography and B-mode ultrasound from 158 patients with breast cancer. We extracted a total of 428 features, consisting of morphologic features from B-mode, and intensity features and gray-level co-occurrence matrix features from the dual modalities, and the optimal subsut of features was selected through least absolute shrinkage and selection operator (Lasso) under the condition of leave-one-out cross validation. We used SVM for the classification of benign and metastatic nodes. Results The sensitivity, specificity, accuracy and Youden's index of the 35 radiomics features selected with Lasso were 86.96%, 85.51%, 86.34% and 72.46%, respectively. Conclusions The radiomics features from dual-modal ultrasound (elastography and B-mode) have demonstrated good performance for classification and have potential to be applied to clinical diagnosis of axillary lymph node metastasis.

Multimodal Ultrasound imaging based diagnosis of liver cancers with a two-stage multi-view learning framework

Abstract: Computer-aided diagnosis (CAD) of liver cancers on contrast-enhanced ultrasound (CEUS) has attracted considerable attention in recent years. The enhancement patterns on CEUS for liver lesions consist of the arterial, portal venous and late phases. Several typical images selected from these three phases can provide reliable information basis for diagnosis of liver lesions. Therefore, we propose to develop a CAD framework for liver cancers with only one B-mode image and three typical CEUS images selected from three enhancement patterns, which simulates the clinical diagnosis mode of radiologists. Moreover, a framework of two-stage multi-view learning (TS-MVL) is proposed to perform both feature-level and classifier-level MVL for the diagnosis of liver cancers with multimodal ultrasound images. We propose to apply the nonlinear kernel matrix (NKM) algorithm to effectively fuse the features of multimodal ultrasound images, and then perform the multiple kernel boosting (MKB) algorithm to promote the predictive performance of multiple classifiers according to multi-view features. The experimental results indicate that the proposed algorithm outperforms the commonly used multi-view learning algorithms.

Robust sound event classification with bilinear multi-column ELM-AE and two-stage ensemble learning

Abstract: The automatic sound event classification (SEC) has attracted a growing attention in recent years Feature extraction is a critical factor in SEC system, and the deep neural network (DNN) algorithms have achieved the state-of-the-art performance for SEC The extreme learning machine-based auto-encoder (ELM-AE) is a new deep learning algorithm, which has both an excellent representation performance and very fast training procedure However, ELM-AE suffers from the problem of unstability In this work, a bilinear multi-column ELM-AE (B-MC-ELM-AE) algorithm is proposed to improve the robustness, stability, and feature representation of the original ELM-AE, which is then applied to learn feature representation of sound signals Moreover, a B-MC-ELM-AE and two-stage ensemble learning (TsEL)-based feature learning and classification framework is then developed to perform the robust and effective SEC The experimental results on the Real World Computing Partnership Sound Scene Database show that the proposed SEC framework outperforms the state-of-the-art DNN algorithm

Elevated hardness of peripheral gland on real-time elastography is an independent marker for high-risk prostate cancers.

Abstract: To examine the role of quantitative real-time elastography (RTE) features on differentiation between high-risk prostate cancer (PCA) and non-high-risk prostatic diseases in the initial transperineal biopsy setting. We retrospectively included 103 patients with suspicious PCA who underwent both RTE and initial transperineal prostate biopsy. Patients were grouped into high-risk and non-high-risk categories according to the D’Amico’s risk stratification. With computer assistance based on MATLAB programming, three features were extracted from RTE, i.e., the median hardness within peripheral gland (PG) (H med), the ratio of the median hardness within PG to that outside PG (H ratio), and the ratio of the hard area within PG to the total PG area (H ar). A multiple regression model incorporating an RTE feature, age, transrectal ultrasound finding, and prostate volume was used to identify markers for high-risk PCA. Forty-seven patients (45.6%) were diagnosed with PCA and 34 (33.0%) were diagnosed with high-risk PCA. Three RTE features were all statistically higher in high-risk PCA than in non-high-risk diseases (p < 0.001), indicating that the PGs in high-risk PCA patients were harder than those in non-high-risk patients. A high H ratio, high age, and low prostate volume were found to be independent markers for PCAs (p < 0.05), among which the high H ratio was the only independent marker for high-risk PCAs (p = 0.012). When predicting high-risk PCAs, the multiple regression achieved an area under receiver operating characteristic curve of 0.755, sensitivity of 73.5%, and specificity of 71.0%. The elevated hardness of PG identified high-risk PCA and served as an independent marker of high-risk PCA. As a non-invasive imaging modality, the RTE could be potentially used in routine clinical practice for the detection of high-risk PCA to decrease unnecessary biopsies and reduce overtreatment.

Age groups related glioblastoma study based on radiomics approach.

Abstract: Glioblastoma is the most aggressive malignant brain tumor with poor prognosis. Radiomics is a newly emerging and promising technique to reveal the complex relationships between high-throughput medical image features and deep information of disease including pathology, biomarkers and genomics. An approach was developed to investigate the internal relationship between magnetic resonance imaging (MRI) features and the age-related origins of glioblastomas based on a quantitative radiomics method. A fully automatic image segmentation method was applied to segment the tumor regions from three dimensional MRI images. 555 features were then extracted from the image data. By analyzing large numbers of quantitative image features, some predictive and prognostic information could be obtained by the radiomics approach. 96 patients diagnosed with glioblastoma pathologically have been divided into two age groups (<45 and ≥45 years old). As expected, there are 101 features showing the consistency with the age gr...

Image Stitching for Achilles Tendon Ultrasonography Using Particle Swarm Optimization of Boundary Vector Field Based Similarity and Log-Domain Diffeomorphic Registration

Abstract: Results: Simulation experiments of non-rigid registration on 20 image pairs show that the horizontal, vertical and distance registration errors of our method are 1.422, 0.628, and 1.616 pixel, respectively. Non-rigid registration experiments on 11 pairs of real clinical US images demonstrate that the horizontal, vertical and distance registration errors are 1.514, 1.205, and 1.928 pixel, respectively. Registration results on both synthetic images and real clinical images by our method exceeded those of other four methods. After image blending on multiple AT sections of US scans, panoramas of AT are generated for dual modality visualization of both B-mode and sonoelastography.

Intensity inhomogeneity compensation and tissue segmentation for magnetic resonance imaging with noise-suppressed multiplicative intrinsic component optimization

Abstract: Magnetic resonance (MR) images suffer from intensity inhomogeneity. Segmentation-based approaches can simultaneously achieve both intensity inhomogeneity compensation (IIC) and tissue segmentation for MR images with little noise, but they often fail for images polluted by severe noise. Here, we propose a noise-robust algorithm named noise-suppressed multiplicative intrinsic component optimization (NSMICO) for simultaneous IIC and tissue segmentation. Considering the spatial characteristics in an image, an adaptive nonlocal means filtering term is incorporated into the objective function of NSMICO to decrease image deterioration due to noise. Then, a fuzzy local factor term utilizing the spatial and gray-level relationship among local pixels is embedded into the objective function to reach a balance between noise suppression and detail preservation. Experimental results on synthetic natural and MR images with various levels of intensity inhomogeneity and noise, as well as in vivo clinical MR images, have demonstrated the effectiveness of the NSMICO and its superiority to three competing approaches. The NSMICO could be potentially valuable for MR image IIC and tissue segmentation.