Learning invariant features of tumor signatures

Abstract

We present a novel method for automated learning of features from unlabeled image patches for classification of tumor architecture. In contrast to manually designed feature detectors (e.g., Gabor basis function), the proposed method utilizes independent subspace analysis to reconstruct a natural representation. Learning is described as a two-layer network with non-linear responses, where the second layer represents subspace structures. The technique is applied to tissue sections for characterizing necrosis, apoptotic, and viable regions of Glioblastoma Multifrome (GBM) from TCGA dataset. We show that the performance of this method is better than expert designed representation, therefore, promising a wider application of self-learning strategies for tissue characterization.Index Terms-subspace learning, tumor architecture, apoptotic and necrotic signatures

Deep convolutional activation features for large scale Brain Tumor histopathology image classification and segmentation

“…Expert designed features include morphometric features [14], fractal features [15], texture features [16] and object-like features [17]. However, study [18] has pointed out that features learned by a two-layer network with non-linear responses using unlabeled image patches are superior to expert designed representations when it comes to histopathology images. Nayak [19] introduces sparse features learning using the restricted Boltzmann machine (RBM) to describe histopathology features in GBM and clear cell kidney carcinoma (KIRC).…”

Deep learning of feature representation with multiple instance learning for medical image analysis

“…The main methods consists of manually feature design, supervised feature learning, and unsupervised feature learning. Boucheron [18] and Chang [19] focus on manual feature design while Le [20] focuses on unsupervised feature learning. Boucheron et al [18] exploited segmentation results of cell nuclei as features to improve the classification accuracy in histopathology images of breast cancer.…”

“…Chang et al [19] presented nuclear level morphometric features at various locations and scales within the spatial pyramid matching to classify tumor histopathology images. Le et al [20] proposed a two-layer network with nonlinear responses to automatically learn features in histopathology tumor images. In our work, we compared the three main methods on a colon histopathology dataset.…”

“…Deep learning as classifiers are used in acoustic emotion recognition [21] and object classes in ImageNet [22]. Deep learning can be used in feature learning including supervised [9] and unsupervised [20]. In our work, we attempted deep learning of feature representation with MIL to classify colon histopathology images.…”

A Convolutional Neural Network approach for classifying leukocoria

“…The first layer of a CNN learns filters similar in concept to the features that are learned by RISA [5], which has been used in the analysis of tumor signatures and outperforms the best known expert-designed feature detector for that task [6].…”