Aydın Ulaş

Dissimilarity-based detection of schizophrenia

We propose to approach the detection of patients affected by schizophrenia by means of dissimilarity-based classification techniques applied to brain magnetic resonance images. Instead of working with features directly, pairwise distances between expert delineated regions of interest (ROIs) are considered as representations based on which learning and classification can be performed. Experiments were carried out on a set of 64 patients and60 controls and several pairwise dissimilarity measurements have been analyzed. We demonstrate that good results are possible and especially significant improvements can be obtained when combining over different ROIs and different distance measures. The lowest error rate obtained is 0.210.

Combining information theoretic kernels with generative embeddings for classification

Classical approaches to learn classifiers for structured objects (e.g., images, sequences) use generative models in a standard Bayesian framework. To exploit the state-of-the-art performance of discriminative learning, while also taking advantage of generative models of the data, generative embeddings have been recently proposed as a way of building hybrid discriminative/generative approaches. A generative embedding is a mapping, induced by a generative model (usually learned from data), from the object space into a fixed dimensional space, adequate for discriminative classifier learning. Generative embeddings have been shown to often outperform the classifiers obtained directly from the generative models upon which they are built. Using a generative embedding for classification involves two main steps: (i) defining and learning a generative model and using it to build the embedding; (ii) discriminatively learning a (maybe kernel) classifier with the embedded data. The literature on generative embeddings is essentially focused on step (i), usually taking some standard off-the-shelf tool for step (ii). Here, we adopt a different approach, by focusing also on the discriminative learning step. In particular, we exploit the probabilistic nature of generative embeddings, by using kernels defined on probability measures; in particular we investigate the use of a recent family of non-extensive information theoretic kernels on the top of different generative embeddings. We show, in different medical applications that the approach yields state-of-the-art performance.

Dissimilarity-Based Detection of Schizophrenia

We propose to approach the detection of patients affected by schizophrenia by means of dissimilarity-based classification techniques applied to brain magnetic resonance images. Instead of working with features directly, pairwise distances between expert delineated regions of interest (ROIs) are considered as representations based on which learning and classification can be performed. Experiments were carried out on a set of 64 patients and60 controls and several pairwise dissimilarity measurements have been analyzed. We demonstrate that good results are possible and especially significant improvements can be obtained when combining over different ROIs and different distance measures. The lowest error rate obtained is 0.210.

Biomarker Evaluation by Multiple Kernel Learning for Schizophrenia Detection

In this paper, we use the promising paradigm of Multiple Kernel Learning (MKL) to challenge the problem of biomarker evaluation for schizophrenia detection. We use eight different Regions of Interest (ROIs) extracted from Magnetic Resonance Images (MRIs). For each region we evaluate both tissue and geometric properties. We show that with MKL we not only obtain more accurate classifiers than using single source support vector machines (SVMs), feature concatenation and kernel averaging but also we evaluate the relevance of the brain biomarkers in predicting this disease. On a data set of 50 patients and 50 healthy controls we can achieve an increase of 7% accuracy compared to standard methods. Moreover, we are able to quantify the importance of each source of information by highlighting the synergies between the involved brain characteristics.

Renal cancer cell classification using generative embeddings and information theoretic kernels

In this paper, we propose a hybrid generative/discriminative classification scheme and apply it to the detection of renal cell carcinoma (RCC) on tissue microarray (TMA) images. In particular we use probabilistic latent semantic analysis (pLSA) as a generative model to perform generative embedding onto the free energy score space (FESS). Subsequently, we use information theoretic kernels on these embeddings to build a kernel based classifier on the FESS. We compare our results with support vector machines based on standard linear kernels and RBF kernels; and with the nearest neighbor (NN) classifier based on the Mahalanobis distance using a diagonal covariance matrix. We conclude that the proposed hybrid approach achieves higher accuracy, revealing itself as a promising approach for this class of problems.

Multimodal schizophrenia detection by multiclassification analysis

We propose a multiclassification analysis to evaluate the relevance of different factors in schizophrenia detection. Several Magnetic Resonance Imaging (MRI) scans of brains are acquired from two sensors: morphological and diffusion MRI. Moreover, 14 Region Of Interests (ROIs) are available to focus the analysis on specific brain subparts. All information is combined to train three types of classifiers to distinguish between healthy and unhealthy subjects. Our contribution is threefold: (i) the classification accuracy improves when multiple factors are taken into account; (ii) proposed procedure allows the selection of a reduced subset of ROIs, and highlights the synergy between the two modalities; (iii) correlation analysis is performed for every ROI and modality to measure the information overlap using the correlation coefficient in the context of schizophrenia classification. We see that we achieve 85.96 % accuracy when we combine classifiers from both modalities, whereas the highest performance of a single modality is 78.95 %.

Hybrid generative-discriminative nucleus classification of renal cell carcinoma

In this paper, we propose to use advanced classification techniques with shape features for nuclei classification in tissue microarray images of renal cell carcinoma. Our aim is to improve the classification accuracy in distinguishing between healthy and cancerous cells. The approach is inspired by natural language processing: several features are extracted from the automatically segmented nuclei and quantized to visual words, and their co-occurrences are encoded as visual topics. To this end, a generative model, the probabilistic Latent Semantic Analysis (pLSA) is learned from quantized shape descriptors (visual words). Finally, we extract from the learned models a generative score, that is used as input for new classifiers, defining a hybrid generative-discriminative classification algorithm. We compare our results with the same classifiers on the feature set to assess the increase of accuracy when we apply pLSA. We demonstrate that the feature space created using pLSA achieves better accuracies than the original feature space.

A multiple Kernel learning algorithm for cell nucleus classification of renal cell carcinoma

We consider a Multiple Kernel Learning (MKL) framework for nuclei classification in tissue microarray images of renal cell carcinoma. Several features are extracted from the automatically segmented nuclei and MKL is applied for classification. We compare our results with an incremental version of MKL, support vector machines with single kernel (SVM) and voting. We demonstrate that MKL inherently combines information from different input spaces and creates statistically significantly more accurate classifiers than SVMs and voting for renal cell carcinoma detection.

Analysis of Brain Magnetic Resonance (MR) Scans for the Diagnosis of Mental Illness

We address the problem of schizophrenia detection by analyzing magnetic resonance imaging (MRI). In general, mental illness like schizophrenia or bipolar disorders are traditionally diagnosed by self-reports and behavioral observations. A new trend in neuroanatomical research consists of using MRI images to find possible connections between cognitive impairments and neuro-physiological abnormalities. Indeed, brain imaging techniques are appealing to provide a non-invasive diagnostic tool for mass analyses and early diagnoses. The problem is challenging due to the heterogeneous behavior of the disease and up to now, although the literature is large in this field, there is not a consolidated framework to deal with it. In this context, advanced pattern recognition and machine learning techniques can be useful to improve the automatization of the involved procedures and the characterization of mental illnesses with specific and detectable brain abnormalities. In this book, we have exploited similarity-based pattern recognition techniques to further improve brain classification problem by employing the algorithms developed in the other chapters of this book. (This chapter is based on previous works (Castellani et al. in Proceedings of the International Conference on Medical Image Computing and Computer-Assisted Intervention, MICCAI’11, vol. 6892, pp. 426–433, 2011; Gonen et al. in Proceedings of the International Workshop on Similarity-Based Pattern Analysis, SIMBAD’11, vol. 7005, pp. 250–260, 2011; Ulas et al. in Proceedings of the Iberoamerican Congress on Pattern Recognition, CIARP’11, vol. 7042, pp. 491–498, 2011; in IAPR International Conference on Pattern Recognition in Bioinformatics, PRIB’11, vol. 7036, pp. 306–317, 2011; and in Int. J. Imaging Syst. Technol. 21(2):179–192, 2011) by the authors and contains text, equations and experimental results taken from these papers.)

A Localized MKL Method for Brain Classification with Known Intra-class Variability

Automatic decisional systems based on pattern classification methods are becoming very important to support medical diagnosis. In general, the overall objective is to classify between healthy subjects and patients affected by a certain disease. To reach this aim, significant efforts have been spent in finding reliable biomarkers which are able to robustly discriminate between the two populations (i.e., patients and controls). However, in real medical scenarios there are many factors, like the gender or the age, which make the source data very heterogeneous. This introduces a large intra-class variation by affecting the performance of the classification procedure. In this paper we exploit how to use the knowledge on heterogeneity factors to improve the classification accuracy. We propose a Clustered Localized Multiple Kernel Learning (CLMKL) algorithm by encoding in the classication model the information on the clusters of apriory known stratifications.