Dimitrios K. Iakovidis

Reduction of capsule endoscopy reading times by unsupervised image mining

The screening of the small intestine has become painless and easy with wireless capsule endoscopy (WCE) that is a revolutionary, relatively non-invasive imaging technique performed by a wireless swallowable endoscopic capsule transmitting thousands of video frames per examination. The average time required for the visual inspection of a full 8-h WCE video ranges from 45 to 120min, depending on the experience of the examiner. In this paper, we propose a novel approach to WCE reading time reduction by unsupervised mining of video frames. The proposed methodology is based on a data reduction algorithm which is applied according to a novel scheme for the extraction of representative video frames from a full length WCE video. It can be used either as a video summarization or as a video bookmarking tool, providing the comparative advantage of being general, unbounded by the finiteness of a training set. The number of frames extracted is controlled by a parameter that can be tuned automatically. Comprehensive experiments on real WCE videos indicate that a significant reduction in the reading times is feasible. In the case of the WCE videos used this reduction reached 85% without any loss of abnormalities.

Fusion of fuzzy statistical distributions for classification of thyroid ultrasound patterns

OBJECTIVEnThis paper proposes a novel approach for thyroid ultrasound pattern representation. Considering that texture and echogenicity are correlated with thyroid malignancy, the proposed approach encodes these sonographic features via a noise-resistant representation. This representation is suitable for the discrimination of nodules of high malignancy risk from normal thyroid parenchyma.nnnMATERIALS AND METHODSnThe material used in this study includes a total of 250 thyroid ultrasound patterns obtained from 75 patients in Greece. The patterns are represented by fused vectors of fuzzy features. Ultrasound texture is represented by fuzzy local binary patterns, whereas echogenicity is represented by fuzzy intensity histograms. The encoded thyroid ultrasound patterns are discriminated by support vector classifiers.nnnRESULTSnThe proposed approach was comprehensively evaluated using receiver operating characteristics (ROCs). The results show that the proposed fusion scheme outperforms previous thyroid ultrasound pattern representation methods proposed in the literature. The best classification accuracy was obtained with a polynomial kernel support vector machine, and reached 97.5% as estimated by the area under the ROC curve.nnnCONCLUSIONSnThe fusion of fuzzy local binary patterns and fuzzy grey-level histogram features is more effective than the state of the art approaches for the representation of thyroid ultrasound patterns and can be effectively utilized for the detection of nodules of high malignancy risk in the context of an intelligent medical system.

M3G: Maximum Margin Microarray Gridding

BackgroundComplementary DNA (cDNA) microarrays are a well established technology for studying gene expression. A microarray image is obtained by laser scanning a hybridized cDNA microarray, which consists of thousands of spots representing chains of cDNA sequences, arranged in a two-dimensional array. The separation of the spots into distinct cells is widely known as microarray image gridding.MethodsIn this paper we propose M3G, a novel method for automatic gridding of cDNA microarray images based on the maximization of the margin between the rows and the columns of the spots. Initially the microarray image rotation is estimated and then a pre-processing algorithm is applied for a rough spot detection. In order to diminish the effect of artefacts, only a subset of the detected spots is selected by matching the distribution of the spot sizes to the normal distribution. Then, a set of grid lines is placed on the image in order to separate each pair of consecutive rows and columns of the selected spots. The optimal positioning of the lines is determined by maximizing the margin between these rows and columns by using a maximum margin linear classifier, effectively facilitating the localization of the spots.ResultsThe experimental evaluation was based on a reference set of microarray images containing more than two million spots in total. The results show that M3G outperforms state of the art methods, demonstrating robustness in the presence of noise and artefacts. More than 98% of the spots reside completely inside their respective grid cells, whereas the mean distance between the spot center and the grid cell center is 1.2 pixels.ConclusionsThe proposed method performs highly accurate gridding in the presence of noise and artefacts, while taking into account the input image rotation. Thus, it provides the potential of achieving perfect gridding for the vast majority of the spots.

TND: A Thyroid Nodule Detection System for Analysis of Ultrasound Images and Videos

In this paper, we present a computer-aided-diagnosis (CAD) system prototype, named TND (Thyroid Nodule Detector), for the detection of nodular tissue in ultrasound (US) thyroid images and videos acquired during thyroid US examinations. The proposed system incorporates an original methodology that involves a novel algorithm for automatic definition of the boundaries of the thyroid gland, and a novel approach for the extraction of noise resilient image features effectively representing the textural and the echogenic properties of the thyroid tissue. Through extensive experimental evaluation on real thyroid US data, its accuracy in thyroid nodule detection has been estimated to exceed 95%. These results attest to the feasibility of the clinical application of TND, for the provision of a second more objective opinion to the radiologists by exploiting image evidences.

Unsupervised SVM-based gridding for DNA microarray images

This paper presents a novel method for unsupervised DNA microarray gridding based on support vector machines (SVMs). Each spot is a small region on the microarray surface where chains of known DNA sequences are attached. The goal of microarray gridding is the separation of the spots into distinct cells. The positions of the spots on a DNA microarray image are first detected using image analysis operations and then a set of soft-margin linear SVM classifiers is used to estimate the optimal layout of the grid lines in the image. Each grid line is the separating line produced by one of the SVM classifiers, which maximizes the margin between two consecutive rows or columns of spots. The classifiers are trained using the spot locations as training vectors. The proposed method was evaluated on reference microarray images containing more than two million spots in total. The results illustrate its robustness in the presence of artifacts, noise and weakly expressed spots, as well as image rotation. The comparison to state of the art methods for microarray gridding reveals the superior performance of the proposed method. In 96.4% of the cases, the spots reside completely inside their respective grid cells.

Automatic lesion detection in wireless capsule endoscopy — A simple solution for a complex problem

Wireless capsule endoscopy (WCE) is performed with a swallowable miniature optical endoscope which transmits color images wirelessly during its journey in the gastrointestinal tract. In this paper we present a computationally efficient and effective approach to cope with automatic detection of possible abnormalities in the WCE videos and consequently with the reduction of the time required for the WCE inspection. It involves automatic detection of salient points based on color information and supervised classification of simple color vectors extracted from the neighborhood of each point. The experiments performed aim to determine the optimal color space components for feature extraction, and identification of abnormalities. Main advantages of this approach are its computational efficiency, its sensitivity to detect small lesions, and its generality. The results obtained from experimentation with a dataset with various types of abnormalities and non-ideal normal frames, approximate 0.9 in terms of the area under receiver operating characteristic (ROC).

Mining Patterns of Lung Infections in Chest Radiographs

Chest radiography is a reference standard and the initial diagnostic test performed in patients who present with signs and symptoms suggesting a pulmonary infection. The most common radiographic manifestation of bacterial pulmonary infections is foci of consolidation. These are visible as bright shadows interfering with the interior lung intensities. The discovery and the assessment of bacterial infections in chest radiographs is a challenging computational task. It has been limitedly addressed as it is subject to image quality variability, content diversity, and deformability of the depicted anatomic structures. In this paper, we propose a novel approach to the discovery of consolidation patterns in chest radiographs. The proposed approach is based on non-negative matrix factorization (NMF) of statistical intensity signatures characterizing the densities of the depicted anatomic structures. Its experimental evaluation demonstrates its capability to recover semantically meaningful information from chest radiographs of patients with bacterial pulmonary infections. Moreover, the results reveal its comparative advantage over the baseline fuzzy C-means clustering approach.

Blood detection in wireless capsule endoscope images based on salient superpixels.

Wireless capsule endoscopy (WCE) enables screening of the gastrointestinal (GI) tract with a miniature, optical endoscope packed within a small swallowable capsule, wirelessly transmitting color images. In this paper we propose a novel method for automatic blood detection in contemporary WCE images. Blood is an alarming indication for the presence of pathologies requiring further treatment. The proposed method is based on a new definition of superpixel saliency. The saliency of superpixels is assessed upon their color, enabling the identification of image regions that are likely to contain blood. The blood patterns are recognized by their color features using a supervised learning machine. Experiments performed on a public dataset using automatically selected first-order statistical features from various color components indicate that the proposed method outperforms state-of-the-art methods.

Comparative assessment of feature extraction methods for visual odometry in wireless capsule endoscopy

Wireless capsule endoscopy (WCE) enables the non-invasive examination of the gastrointestinal (GI) tract by a swallowable device equipped with a miniature camera. Accurate localization of the capsule in the GI tract enables accurate localization of abnormalities for medical interventions such as biopsy and polyp resection; therefore, the optimization of the localization outcome is important. Current approaches to endoscopic capsule localization are mainly based on external sensors and transit time estimations. Recently, we demonstrated the feasibility of capsule localization based-entirely-on visual features, without the use of external sensors. This technique relies on a motion estimation algorithm that enables measurements of the distance and the rotation of the capsule from the acquired video frames. Towards the determination of an optimal visual feature extraction technique for capsule motion estimation, an extensive comparative assessment of several state-of-the-art techniques, using a publicly available dataset, is presented. The results show that the minimization of the localization error is possible at the cost of computational efficiency. A localization error of approximately one order of magnitude higher than the minimal one can be considered as compromise for the use of current computationally efficient feature extraction techniques.

Ontology-Based automatic image annotation exploiting generalized qualitative spatial semantics

Ontologies provide a formal approach to knowledge representation suitable for digital content annotation. In the context of image annotation a variety of ontology-based tools has been proposed. Most of them enable manual annotation of the images with higher level concepts whereas many of them are capable of formally representing low-level features as well. However, they either consider specific, usually quantitative, representations of the low-level features, or spatial semantics limited to 2D/3D image spaces. In this paper we propose a novel ontology-based methodology for automatic image annotation that exploits generalized qualitative spatial relations between objects, given an image domain. To represent knowledge for the spatial arrangements, we have implemented an ontology that models spatial relations in multi-dimensional vector spaces. The application of the proposed methodology is demonstrated for automatic annotation of segmented objects in chest radiographs.