An unsupervised feature learning framework for basal cell carcinoma image analysis

Computer assisted diagnosis of basal cell carcinoma using Z-transform features

“…Our work extends the method proposed in [30] in the sense that we also employ features that can best reconstruct the input image, but it differs from that work in two ways: 1) we use frequency domain features instead of spatial domain ones introduced in [30]. Specially, histopathological images are classified using Z-transform coefficients which will be shown to have better discriminative power with respect to the conventional Fourier transform coefficients and those used in [30].…”

“…The problem of basal cell carcinoma detection is addressed in [26][27][28][29][30]. In [26], cells nuclei are segmented by maximally stable extreme regions (MSER) approach and then, color descriptors are extracted from the segmented regions at different scales.…”

“…Authors in [27] proposed a graph-based method along with a Bayesian strategy to identify the regions of interest in histopathological images of basal cell carcinoma. Unsupervised feature learning (UFL) methods are used in [28][29][30] for basal cell carcinoma detection. Specially, authors in [30] employed three kinds of UFL approaches (sparse auto encoders, reconstruct independent component analysis and topographic independent component analysis) along with the convolutional representation strategy to classify basal cell carcinoma against normal tissue.…”

“…Unsupervised feature learning (UFL) methods are used in [28][29][30] for basal cell carcinoma detection. Specially, authors in [30] employed three kinds of UFL approaches (sparse auto encoders, reconstruct independent component analysis and topographic independent component analysis) along with the convolutional representation strategy to classify basal cell carcinoma against normal tissue. First, the system learns a set of feature detectors from the image blocks of the samples in the training set.…”

A survey on computational intelligence approaches for predictive modeling in prostate cancer

“…Many types of Deep Neural Networks exist, some of which are the 325 Deep Boltzmann Machines (Salakhutdinov & Hinton, 2009), the Restricted Deep Boltzmann machine (Hinton & Sejnowski, 1986), and the Convolutional Deep Belief Network (Lee et al, 2009). These methods have dramatically improved state-of-the-art natural language processing (Mikolov et al, 2013), computer vision (Ciresan et al, 2012), as well as many other applications such as drug 330 discovery and genomics (LeCun et al, 2015), and the analysis carcinoma images (Arevalo et al, 2015a). Convolutional neural networks have been applied for classifying mass lesions following mammography (Arevalo et al, 2015b) …”