Classification of Tumor Histology via Morphometric Context

Clinical significance and molecular annotation of cellular morphometric subtypes in lower-grade gliomas discovered by machine learning

BACKGROUND

Lower-grade gliomas (LGG) are heterogeneous diseases by clinical, histological, and molecular criteria. We aimed to personalize the diagnosis and therapy of LGG patients by developing and validating robust cellular morphometric subtypes (CMS) and to uncover the molecular signatures underlying these subtypes.

METHODS

Cellular morphometric biomarkers (CMBs) were identified with artificial intelligence technique from TCGA-LGG cohort. Consensus clustering was used to define CMS. Survival analysis was performed to assess the clinical impact of CMBs and CMS. A nomogram was constructed to predict 3- and 5-year overall survival (OS) of LGG patients. Tumor mutational burden (TMB) and immune cell infiltration between subtypes were analyzed using the Mann-Whitney U test. The double-blinded validation for important immunotherapy-related biomarkers was executed using immunohistochemistry (IHC).

RESULTS

We developed a machine learning (ML) pipeline to extract CMBs from whole-slide images of tissue histology; identifying and externally validating robust CMS of LGGs in multicenter cohorts. The subtypes had independent predicted OS across all three independent cohorts. In the TCGA-LGG cohort, patients within the poor-prognosis subtype responded poorly to primary and follow-up therapies. LGGs within the poor-prognosis subtype were characterized by high mutational burden, high frequencies of copy number alterations, and high levels of tumor-infiltrating lymphocytes and immune checkpoint genes. Higher levels of PD-1/PD-L1/CTLA-4 were confirmed by IHC staining. In addition, the subtypes learned from LGG demonstrate translational impact on glioblastoma (GBM).

CONCLUSIONS

We developed and validated a framework (CMS-ML) for CMS discovery in LGG associated with specific molecular alterations, immune microenvironment, prognosis, and treatment response.

Lung cancer subtype classification using histopathological images based on weakly supervised multi-instance learning

Objective.Subtype classification plays a guiding role in the clinical diagnosis and treatment of non-small-cell lung cancer (NSCLC). However, due to the gigapixel of whole slide images (WSIs) and the absence of definitive morphological features, most automatic subtype classification methods for NSCLC require manually delineating the regions of interest (ROIs) on WSIs.Approach.In this paper, a weakly supervised framework is proposed for accurate subtype classification while freeing pathologists from pixel-level annotation. With respect to the characteristics of histopathological images, we design a two-stage structure with ROI localization and subtype classification. We first develop a method called multi-resolution expectation-maximization convolutional neural network (MR-EM-CNN) to locate ROIs for subsequent subtype classification. The EM algorithm is introduced to select the discriminative image patches for training a patch-wise network, with only WSI-wise labels available. A multi-resolution mechanism is designed for fine localization, similar to the coarse-to-fine process of manual pathological analysis. In the second stage, we build a novel hierarchical attention multi-scale network (HMS) for subtype classification. HMS can capture multi-scale features flexibly driven by the attention module and implement hierarchical features interaction.Results.Experimental results on the 1002-patient Cancer Genome Atlas dataset achieved an AUC of 0.9602 in the ROI localization and an AUC of 0.9671 for subtype classification.Significance.The proposed method shows superiority compared with other algorithms in the subtype classification of NSCLC. The proposed framework can also be extended to other classification tasks with WSIs.

Weakly Supervised Deep Learning for Whole Slide Lung Cancer Image Analysis

Histopathology image analysis serves as the gold standard for cancer diagnosis. Efficient and precise diagnosis is quite critical for the subsequent therapeutic treatment of patients. So far, computer-aided diagnosis has not been widely applied in pathological field yet as currently well-addressed tasks are only the tip of the iceberg. Whole slide image (WSI) classification is a quite challenging problem. First, the scarcity of annotations heavily impedes the pace of developing effective approaches. Pixelwise delineated annotations on WSIs are time consuming and tedious, which poses difficulties in building a large-scale training dataset. In addition, a variety of heterogeneous patterns of tumor existing in high magnification field are actually the major obstacle. Furthermore, a gigapixel scale WSI cannot be directly analyzed due to the immeasurable computational cost. How to design the weakly supervised learning methods to maximize the use of available WSI-level labels that can be readily obtained in clinical practice is quite appealing. To overcome these challenges, we present a weakly supervised approach in this article for fast and effective classification on the whole slide lung cancer images. Our method first takes advantage of a patch-based fully convolutional network (FCN) to retrieve discriminative blocks and provides representative deep features with high efficiency. Then, different context-aware block selection and feature aggregation strategies are explored to generate globally holistic WSI descriptor which is ultimately fed into a random forest (RF) classifier for the image-level prediction. To the best of our knowledge, this is the first study to exploit the potential of image-level labels along with some coarse annotations for weakly supervised learning. A large-scale lung cancer WSI dataset is constructed in this article for evaluation, which validates the effectiveness and feasibility of the proposed method. Extensive experiments demonstrate the superior performance of our method that surpasses the state-of-the-art approaches by a significant margin with an accuracy of 97.3%. In addition, our method also achieves the best performance on the public lung cancer WSIs dataset from The Cancer Genome Atlas (TCGA). We highlight that a small number of coarse annotations can contribute to further accuracy improvement. We believe that weakly supervised learning methods have great potential to assist pathologists in histology image diagnosis in the near future.

Localization of Nuclei in Breast Cancer Using Whole Slide Imaging System Supported by Morphological Features and Shape Formulas

PURPOSE

Cancer rates are exponentially increasing worldwide and over 15 million new cases are expected in the year 2020 according to the World Cancer Report. To support the clinical diagnosis of the disease, recent technical advancements in digital microscopy have been achieved to reduce the cost and increase the efficiency of the process. Food and Drug Administration (FDA or Agency) has issued the guidelines, in particular, the development of digital whole slide image scanning system. It is very helpful to the computer-aided diagnosis of breast cancer.

METHODS

Whole slide imaging supported by fluorescence, immunohistochemistry, and multispectral imaging concepts. Due to the high dimension of WSI images and computation, it is a challenging task to find the region of interest (ROI) on a malignant sample image. The unsupervised machine learning and quantitative analysis of malignant sample images are supported by morphological features and shape formulas to find the correct region of interest. Due to computational limitations, it starts to work on small patches, integrate the results, and automated localize or detect the ROI. It is also compared to the handcrafted and automated region of interest provided in the ICIAR2018 dataset.

RESULTS

A total of 10 hematoxylins and eosin (H&E) stained malignant breast histology microscopy whole slide image samples are labeled and annotated by two medical experts who are team members of the ICIAR 2018 challenge. After applying the proposed methodology, it is successfully able to localize the malignant patches of WSI sample images and getting the ROI with an average accuracy of 85.5%.

CONCLUSION

With the help of the k-means clustering algorithm, morphological features, and shape formula, it is possible to recognize the region of interest using the whole slide imaging concept.

Whole slide images reflect DNA methylation patterns of human tumors

DNA methylation is an important epigenetic mechanism regulating gene expression and its role in carcinogenesis has been extensively studied. High-throughput DNA methylation assays have been used broadly in cancer research. Histopathology images are commonly obtained in cancer treatment, given that tissue sampling remains the clinical gold-standard for diagnosis. In this work, we investigate the interaction between cancer histopathology images and DNA methylation profiles to provide a better understanding of tumor pathobiology at the epigenetic level. We demonstrate that classical machine learning algorithms can associate the DNA methylation profiles of cancer samples with morphometric features extracted from whole slide images. Furthermore, grouping the genes into methylation clusters greatly improves the performance of the models. The well-predicted genes are enriched in key pathways in carcinogenesis including hypoxia in glioma and angiogenesis in renal cell carcinoma. Our results provide new insights into the link between histopathological and molecular data.

Automatic Detection of Granuloma Necrosis in Pulmonary Tuberculosis Using a Two-Phase Algorithm: 2D-TB

Granuloma necrosis occurs in hosts susceptible to pathogenic mycobacteria and is a diagnostic visual feature of pulmonary tuberculosis (TB) in humans and in super-susceptible Diversity Outbred (DO) mice infected with Mycobacterium tuberculosis. Currently, no published automated algorithms can detect granuloma necrosis in pulmonary TB. However, such a method could reduce variability, and transform visual patterns into quantitative data for statistical and machine learning analyses. Here, we used histopathological images from super-susceptible DO mice to train, validate, and performance test an algorithm to detect regions of cell-poor necrosis. The algorithm, named 2D-TB, works on 2-dimensional histopathological images in 2 phases. In phase 1, granulomas are detected following background elimination. In phase 2, 2D-TB searches within granulomas for regions of cell-poor necrosis. We used 8 lung sections from 8 different super-susceptible DO mice for training and 10-fold cross validation. We used 13 new lung sections from 10 different super-susceptible DO mice for performance testing. 2D-TB reached 100.0% sensitivity and 91.8% positive prediction value. Compared to an expert pathologist, agreement was 95.5% and there was a statistically significant positive correlation for area detected by 2D-TB and the pathologist. These results show the development, validation, and accurate performance of 2D-TB to detect granuloma necrosis.

High-Dimensional Phenotyping Identifies Age-Emergent Cells in Human Mammary Epithelia

Aging is associated with tissue-level changes in cellular composition that are correlated with increased susceptibility to disease. Aging human mammary tissue shows skewed progenitor cell potency, resulting in diminished tumor-suppressive cell types and the accumulation of defective epithelial progenitors. Quantitative characterization of these age-emergent human cell subpopulations is lacking, impeding our understanding of the relationship between age and cancer susceptibility. We conducted single-cell resolution proteomic phenotyping of healthy breast epithelia from 57 women, aged 16–91 years, using mass cytometry. Remarkable heterogeneity was quantified within the two mammary epithelial lineages. Population partitioning identified a subset of aberrant basal-like luminal cells that accumulate with age and originate from age-altered progenitors. Quantification of age-emergent phenotypes enabled robust classification of breast tissues by age in healthy women. This high-resolution mapping highlighted specific epithelial subpopulations that change with age in a manner consistent with increased susceptibility to breast cancer.

•

CyTOF analysis reveals human mammary epithelial heterogeneity with age

•

Age-emergent luminal cells share phenotypes with candidate breast cancer cells of origin

•

Classification models correctly assign tissue samples to their age group

•

Age-related changes are conserved between mammary epithelial tissue and primary cells

Unsupervised Transfer Learning via Multi-Scale Convolutional Sparse Coding for Biomedical Applications

The capabilities of (I) learning transferable knowledge across domains; and (II) fine-tuning the pre-learned base knowledge towards tasks with considerably smaller data scale are extremely important. Many of the existing transfer learning techniques are supervised approaches, among which deep learning has the demonstrated power of learning domain transferrable knowledge with large scale network trained on massive amounts of labeled data. However, in many biomedical tasks, both the data and the corresponding label can be very limited, where the unsupervised transfer learning capability is urgently needed. In this paper, we proposed a novel multi-scale convolutional sparse coding (MSCSC) method, that (I) automatically learns filter banks at different scales in a joint fashion with enforced scale-specificity of learned patterns; and (II) provides an unsupervised solution for learning transferable base knowledge and fine-tuning it towards target tasks. Extensive experimental evaluation of MSCSC demonstrates the effectiveness of the proposed MSCSC in both regular and transfer learning tasks in various biomedical domains.

Deep learning based Nucleus Classification in pancreas histological images

Tumor specimens contain a variety of healthy cells as well as cancerous cells, and this heterogeneity underlies resistance to various cancer therapies. But this problem has not been thoroughly investigated until recently. Meanwhile, technological breakthroughs in imaging have led to an explosion of molecular and cellular profiling data from large numbers of samples, and modern machine learning approaches including deep learning have been shown to produce encouraging results by finding hidden structures and make accurate predictions. In this paper, we propose a Deep learning based Nucleus Classification (DeepNC) approach using paired histopathology and immunofluorescence images (for label), and demonstrate its classification prediction power. This method can solve current issue on discrepancy between genomic- or transcriptomic-based and pathology-based tumor purity estimates by improving histological evaluation. We also explain challenges in training a deep learning model for huge dataset.

Large scale tissue histopathology image classification, segmentation, and visualization via deep convolutional activation features

BACKGROUND

Histopathology image analysis is a gold standard for cancer recognition and diagnosis. Automatic analysis of histopathology images can help pathologists diagnose tumor and cancer subtypes, alleviating the workload of pathologists. There are two basic types of tasks in digital histopathology image analysis: image classification and image segmentation. Typical problems with histopathology images that hamper automatic analysis include complex clinical representations, limited quantities of training images in a dataset, and the extremely large size of singular images (usually up to gigapixels). The property of extremely large size for a single image also makes a histopathology image dataset be considered large-scale, even if the number of images in the dataset is limited.

RESULTS

In this paper, we propose leveraging deep convolutional neural network (CNN) activation features to perform classification, segmentation and visualization in large-scale tissue histopathology images. Our framework transfers features extracted from CNNs trained by a large natural image database, ImageNet, to histopathology images. We also explore the characteristics of CNN features by visualizing the response of individual neuron components in the last hidden layer. Some of these characteristics reveal biological insights that have been verified by pathologists. According to our experiments, the framework proposed has shown state-of-the-art performance on a brain tumor dataset from the MICCAI 2014 Brain Tumor Digital Pathology Challenge and a colon cancer histopathology image dataset.

CONCLUSIONS

The framework proposed is a simple, efficient and effective system for histopathology image automatic analysis. We successfully transfer ImageNet knowledge as deep convolutional activation features to the classification and segmentation of histopathology images with little training data. CNN features are significantly more powerful than expert-designed features.

Integrative Analysis of Cellular Morphometric Context Reveals Clinically Relevant Signatures in Lower Grade Glioma

Integrative analysis based on quantitative representation of whole slide images (WSIs) in a large histology cohort may provide predictive models of clinical outcome. On one hand, the efficiency and effectiveness of such representation is hindered as a result of large technical variations (e.g., fixation, staining) and biological heterogeneities (e.g., cell type, cell state) that are always present in a large cohort. On the other hand, perceptual interpretation/validation of important multivariate phenotypic signatures are often difficult due to the loss of visual information during feature transformation in hyperspace. To address these issues, we propose a novel approach for integrative analysis based on cellular morphometric context, which is a robust representation of WSI, with the emphasis on tumor architecture and tumor heterogeneity, built upon cellular level morphometric features within the spatial pyramid matching (SPM) framework. The proposed approach is applied to The Cancer Genome Atlas (TCGA) lower grade glioma (LGG) cohort, where experimental results (i) reveal several clinically relevant cellular morphometric types, which enables both perceptual interpretation/validation and further investigation through gene set enrichment analysis; and (ii) indicate the significantly increased survival rates in one of the cellular morphometric context subtypes derived from the cellular morphometric context.

When machine vision meets histology: A comparative evaluation of model architecture for classification of histology sections

Classification of histology sections in large cohorts, in terms of distinct regions of microanatomy (e.g., stromal) and histopathology (e.g., tumor, necrosis), enables the quantification of tumor composition, and the construction of predictive models of genomics and clinical outcome. To tackle the large technical variations and biological heterogeneities, which are intrinsic in large cohorts, emerging systems utilize either prior knowledge from pathologists or unsupervised feature learning for invariant representation of the underlying properties in the data. However, to a large degree, the architecture for tissue histology classification remains unexplored and requires urgent systematical investigation. This paper is the first attempt to provide insights into three fundamental questions in tissue histology classification: I. Is unsupervised feature learning preferable to human engineered features? II. Does cellular saliency help? III. Does the sparse feature encoder contribute to recognition? We show that (a) in I, both Cellular Morphometric Feature and features from unsupervised feature learning lead to superior performance when compared to SIFT and [Color, Texture]; (b) in II, cellular saliency incorporation impairs the performance for systems built upon pixel-/patch-level features; and (c) in III, the effect of the sparse feature encoder is correlated with the robustness of features, and the performance can be consistently improved by the multi-stage extension of systems built upon both Cellular Morphmetric Feature and features from unsupervised feature learning. These insights are validated with two cohorts of Glioblastoma Multiforme (GBM) and Kidney Clear Cell Carcinoma (KIRC).

PHENOTYPIC CHARACTERIZATION OF BREAST INVASIVE CARCINOMA VIA TRANSFERABLE TISSUE MORPHOMETRIC PATTERNS LEARNED FROM GLIOBLASTOMA MULTIFORME

Quantitative analysis of whole slide images (WSIs) in a large cohort may provide predictive models of clinical outcome. However, the performance of the existing techniques is hindered as a result of large technical variations (e.g., fixation, staining) and biological heterogeneities (e.g., cell type, cell state) that are always present in a large cohort. Although unsupervised feature learning provides a promising way in learning pertinent features without human intervention, its capability can be greatly limited due to the lack of well-curated examples. In this paper, we explored the transferability of knowledge acquired from a well-curated Glioblastoma Multiforme (GBM) dataset through its application to the representation and characterization of tissue histology from the Cancer Genome Atlas (TCGA) Breast Invasive Carcinoma (BRCA) cohort. Our experimental results reveals two major phenotypic subtypes with statistically significantly different survival curves. Further differential expression analysis of these two subtypes indicates enrichment of genes regulated by NF-kB in response to TNF and genes up-regulated in response to IFNG.

NUCLEI SEGMENTATION VIA SPARSITY CONSTRAINED CONVOLUTIONAL REGRESSION

Automated profiling of nuclear architecture, in histology sections, can potentially help predict the clinical outcomes. However, the task is challenging as a result of nuclear pleomorphism and cellular states (e.g., cell fate, cell cycle), which are compounded by the batch effect (e.g., variations in fixation and staining). Present methods, for nuclear segmentation, are based on human-designed features that may not effectively capture intrinsic nuclear architecture. In this paper, we propose a novel approach, called sparsity constrained convolutional regression (SCCR), for nuclei segmentation. Specifically, given raw image patches and the corresponding annotated binary masks, our algorithm jointly learns a bank of convolutional filters and a sparse linear regressor, where the former is used for feature extraction, and the latter aims to produce a likelihood for each pixel being nuclear region or background. During classification, the pixel label is simply determined by a thresholding operation applied on the likelihood map. The method has been evaluated using the benchmark dataset collected from The Cancer Genome Atlas (TCGA). Experimental results demonstrate that our method outperforms traditional nuclei segmentation algorithms and is able to achieve competitive performance compared to the state-of-the-art algorithm built upon human-designed features with biological prior knowledge.

Automated discrimination of lower and higher grade gliomas based on histopathological image analysis

INTRODUCTION

Histopathological images have rich structural information, are multi-channel in nature and contain meaningful pathological information at various scales. Sophisticated image analysis tools that can automatically extract discriminative information from the histopathology image slides for diagnosis remain an area of significant research activity. In this work, we focus on automated brain cancer grading, specifically glioma grading. Grading of a glioma is a highly important problem in pathology and is largely done manually by medical experts based on an examination of pathology slides (images). To complement the efforts of clinicians engaged in brain cancer diagnosis, we develop novel image processing algorithms and systems to automatically grade glioma tumor into two categories: Low-grade glioma (LGG) and high-grade glioma (HGG) which represent a more advanced stage of the disease.

RESULTS

We propose novel image processing algorithms based on spatial domain analysis for glioma tumor grading that will complement the clinical interpretation of the tissue. The image processing techniques are developed in close collaboration with medical experts to mimic the visual cues that a clinician looks for in judging of the grade of the disease. Specifically, two algorithmic techniques are developed: (1) A cell segmentation and cell-count profile creation for identification of Pseudopalisading Necrosis, and (2) a customized operation of spatial and morphological filters to accurately identify microvascular proliferation (MVP). In both techniques, a hierarchical decision is made via a decision tree mechanism. If either Pseudopalisading Necrosis or MVP is found present in any part of the histopathology slide, the whole slide is identified as HGG, which is consistent with World Health Organization guidelines. Experimental results on the Cancer Genome Atlas database are presented in the form of: (1) Successful detection rates of pseudopalisading necrosis and MVP regions, (2) overall classification accuracy into LGG and HGG categories, and (3) receiver operating characteristic curves which can facilitate a desirable trade-off between HGG detection and false-alarm rates.

CONCLUSION

The proposed method demonstrates fairly high accuracy and compares favorably against best-known alternatives such as the state-of-the-art WND-CHARM feature set provided by NIH combined with powerful support vector machine classifier. Our results reveal that the proposed method can be beneficial to a clinician in effectively separating histopathology slides into LGG and HGG categories, particularly where the analysis of a large number of slides is needed. Our work also reveals that MVP regions are much harder to detect than Pseudopalisading Necrosis and increasing accuracy of automated image processing for MVP detection emerges as a significant future research direction.

Classification of 3D Multicellular Organization in Phase Microscopy for High Throughput Screening of Therapeutic Targets

The current trend in high throughput screening is the utilization of more complex model systems that mimic both structural and functional properties of cellular processes in vivo. In this context, 3D cell culture models have emerged as effective systems to study tumor initiation and cancer behavior, where colony organization represents distinct phenotypic signatures that enable differentiation of cancer cells in culture using phase imaging and in the absence of clinical markers. If the colony organization can be classified into different phenotypes, it will enable rapid drug screening using phase microscopy. In this paper, we propose a novel method based on locality-constrained dictionary learning for the discrimination of aberrant colony organization in phase images, which encodes original SIFT (Scale-Invariant Feature Transform) features into high dimensional sparse codes with locality-preserving landmark points on the nonlinear manifold, and summarizes the sparse features at various locations and scales through spatial pyramid matching for robust representation. Experimental results demonstrate the significant improvement of performance, compared to the state-of-art in the field.

Stacked Predictive Sparse Decomposition for Classification of Histology Sections

Image-based classification of histology sections, in terms of distinct components (e.g., tumor, stroma, normal), provides a series of indices for histology composition (e.g., the percentage of each distinct components in histology sections), and enables the study of nuclear properties within each component. Furthermore, the study of these indices, constructed from each whole slide image in a large cohort, has the potential to provide predictive models of clinical outcome. For example, correlations can be established between the constructed indices and the patients' survival information at cohort level, which is a fundamental step towards personalized medicine. However, performance of the existing techniques is hindered as a result of large technical variations (e.g., variations of color/textures in tissue images due to non-standard experimental protocols) and biological heterogeneities (e.g., cell type, cell state) that are always present in a large cohort. We propose a system that automatically learns a series of dictionary elements for representing the underlying spatial distribution using stacked predictive sparse decomposition. The learned representation is then fed into the spatial pyramid matching framework with a linear support vector machine classifier. The system has been evaluated for classification of distinct histological components for two cohorts of tumor types. Throughput has been increased by using of graphical processing unit (GPU), and evaluation indicates a superior performance results, compared with previous research.

Classification of Histology Sections via Multispectral Convolutional Sparse Coding

Image-based classification of histology sections plays an important role in predicting clinical outcomes. However this task is very challenging due to the presence of large technical variations (e.g., fixation, staining) and biological heterogeneities (e.g., cell type, cell state). In the field of biomedical imaging, for the purposes of visualization and/or quantification, different stains are typically used for different targets of interest (e.g., cellular/subcellular events), which generates multi-spectrum data (images) through various types of microscopes and, as a result, provides the possibility of learning biological-component-specific features by exploiting multispectral information. We propose a multispectral feature learning model that automatically learns a set of convolution filter banks from separate spectra to efficiently discover the intrinsic tissue morphometric signatures, based on convolutional sparse coding (CSC). The learned feature representations are then aggregated through the spatial pyramid matching framework (SPM) and finally classified using a linear SVM. The proposed system has been evaluated using two large-scale tumor cohorts, collected from The Cancer Genome Atlas (TCGA). Experimental results show that the proposed model 1) outperforms systems utilizing sparse coding for unsupervised feature learning (e.g., PSD-SPM [5]); 2) is competitive with systems built upon features with biological prior knowledge (e.g., SMLSPM [4]).

Stacked Predictive Sparse Coding for Classification of Distinct Regions of Tumor Histopathology

Image-based classification of tissue histology, in terms of distinct histopathology (e.g., tumor or necrosis regions), provides a series of indices for tumor composition. Furthermore, aggregation of these indices from each whole slide image (WSI) in a large cohort can provide predictive models of clinical outcome. However, the performance of the existing techniques is hindered as a result of large technical variations (e.g., fixation, staining) and biological heterogeneities (e.g., cell type, cell state) that are always present in a large cohort. We suggest that, compared with human engineered features widely adopted in existing systems, unsupervised feature learning is more tolerant to batch effect (e.g., technical variations associated with sample preparation) and pertinent features can be learned without user intervention. This leads to a novel approach for classification of tissue histology based on unsupervised feature learning and spatial pyramid matching (SPM), which utilize sparse tissue morphometric signatures at various locations and scales. This approach has been evaluated on two distinct datasets consisting of different tumor types collected from The Cancer Genome Atlas (TCGA), and the experimental results indicate that the proposed approach is (i) extensible to different tumor types; (ii) robust in the presence of wide technical variations and biological heterogeneities; and (iii) scalable with varying training sample sizes.