Vertebrae, IVD and spinal canal boundary extraction on MRI, utilizing CT-trained active shape models.

PURPOSE: Vertebrae, intervertebral disc (IVD) and spinal canal (SC) displacements are in the root of several spinal cord pathologies. The localization and boundary extraction of these structures, along with the quantification of their displacements, provide valuable clues for assessing each pathological condition. In this work, we propose a computational method for boundary extraction of vertebrae, IVD and SC in magnetic resonance images (MRI). METHOD: Vertebrae shape priors derived from computed tomography (CT) images are used to guide vertebrae, IVD and SC boundary extraction in MRI. This strategy is dictated by three considerations: (1) CT is the modality of choice for highlighting solid structures such as vertebrae, (2) vertebrae boundaries indirectly impose constraints on the boundaries of neighbouring structures (IVD and SC), and (3) it can be observed that edges are similarly located in CT and MR images; therefore, gradient profiles and shape priors learned by active shape models (ASMs) from CT are also valid in MRI. RESULTS: Experimental comparisons on two MR image datasets demonstrate that the proposed approach obtains segmentation results, which are comparable to the state of the art. Moreover, the adopted bimodal strategy is validated by demonstrating that CT-derived shape priors lead to more accurate boundary extraction than MRI-derived shape priors, even in the case of MR image applications. CONCLUSION: Unlike existing bimodal methods, the proposed one is not dependent on the availability of CT/MR image pairs, which are not usually acquired from the same patient. In addition, unlike state-of-the-art deep learning-based methods, it is not dependent on large amounts of training data. The proposed method requires a limited amount of user intervention.

A condition-independent framework for the classification of error-related brain activity.

The cognitive processing and detection of errors is important in the adaptation of the behavioral and learning processes. This brain activity is often reflected as distinct patterns of event-related potentials (ERPs) that can be employed in the detection and interpretation of the cerebral responses to erroneous stimuli. However, high-accuracy cross-condition classification is challenging due to the significant variations of the error-related ERP components (ErrPs) between complexity conditions, thus hindering the development of error recognition systems. In this study, we employed support vector machines (SVM) classification methods, based on waveform characteristics of ErrPs from different time windows, to detect correct and incorrect responses in an audio identification task with two conditions of different complexity. Since the performance of the classifiers usually depends on the salience of the features employed, a combination of the sequential forward floating feature selection (SFFS) and sequential forward feature selection (SFS) methods was implemented to detect condition-independent and condition-specific feature subsets. Our framework achieved high accuracy using a small subset of the available features both for cross- and within-condition classification, hence supporting the notion that machine learning techniques can detect hidden patterns of ErrP-based features, irrespective of task complexity while additionally elucidating complexity-related error processing variations. Graphical abstract A schematic of the proposed approach. (a) EEG recordings in an auditory experiment in two conditions of different complexity. (b) Characteristic event related activity feature extraction. (c) Selection of feature vector subsets for easy and hard conditions corresponding to correct (Class1) and incorrect (Class2) responses. (d) Performance for individual and cross-condition classification.

Detection of retinal pigment epithelium detachment from OCT images using multiscale Gaussian filtering.

Macular diseases, including neovascular age-related macular degeneration (nvAMD), are leading causes of irreversible blindness and visual impairment. One prominent feature of nvAMD is the detachment of the retinal pigment epithelium. The aim of this study is to implement an automated method for the segmentation of the pigment epithelial detachment (PED) using optical coherence tomography (OCT). OCT datasets from 8 patients with nvAMD were acquired during multiple sessions. At each session, 17 images with a resolution of 1020 × 640 pixels were obtained. The images were segmented using Gaussian filtering and template matching for the detection of the upper and lower border of the PED, respectively. The results of the method were compared with the ones obtained from the manual segmentation of the images by an expert. Four well-known metrics were used to evaluate the performance of the method with respect to the manual segmentation, resulting in high scores of consistency. Furthermore, the proposed method was also compared with four other well-known methods providing similar or superior performance.

Adaptable pattern recognition system for discriminating Melanocytic Nevi from Malignant Melanomas using plain photography images from different image databases.

OBJECTIVE: The aim of this study was to propose features that evaluate pictorial differences between melanocytic nevus (mole) and melanoma lesions by computer-based analysis of plain photography images and to design a cross-platform, tunable, decision support system to discriminate with high accuracy moles from melanomas in different publicly available image databases. MATERIAL AND METHODS: Digital plain photography images of verified mole and melanoma lesions were downloaded from (i) Edinburgh University Hospital, UK, (Dermofit, 330moles/70 melanomas, under signed agreement), from 5 different centers (Multicenter, 63moles/25 melanomas, publicly available), and from the Groningen University, Netherlands (Groningen, 100moles/70 melanomas, publicly available). Images were processed for outlining the lesion-border and isolating the lesion from the surrounding background. Fourteen features were generated from each lesion evaluating texture (4), structure (5), shape (4) and color (1). Features were subjected to statistical analysis for determining differences in pictorial properties between moles and melanomas. The Probabilistic Neural Network (PNN) classifier, the exhaustive search features selection, the leave-one-out (LOO), and the external cross-validation (ECV) methods were used to design the PR-system for discriminating between moles and melanomas. RESULTS: Statistical analysis revealed that melanomas as compared to moles were of lower intensity, of less homogenous surface, had more dark pixels with intensities spanning larger spectra of gray-values, contained more objects of different sizes and gray-levels, had more asymmetrical shapes and irregular outlines, had abrupt intensity transitions from lesion to background tissue, and had more distinct colors. The PR-system designed by the Dermofit images scored on the Dermofit images, using the ECV, 94.1%, 82.9%, 96.5% for overall accuracy, sensitivity, specificity, on the Multicenter Images 92.0%, 88%, 93.7% and on the Groningen Images 76.2%, 73.9%, 77.8% respectively. CONCLUSION: The PR-system as designed by the Dermofit image database could be fine-tuned to classify with good accuracy plain photography moles/melanomas images of other databases employing different image capturing equipment and protocols.

Computer-based automated estimation of breast vascularity and correlation with breast cancer in DCE-MRI images.

Dynamic contrast enhanced magnetic resonance imaging (DCE-MRI) with gadolinium constitutes one of the most promising protocols for boosting up the sensitivity in breast cancer detection. The aim of this study was twofold: first to design an image processing methodology to estimate the vascularity of the breast region in DCE-MRI images and second to investigate whether the differences in the composition/texture and vascularity of normal, benign and malignant breasts may serve as potential indicators regarding the presence of the disease. Clinical material comprised thirty nine cases examined on a 3.0-T MRI system (SIGNA HDx; GE Healthcare). Vessel segmentation was performed using a custom made modification of the Seeded Region Growing algorithm that was designed in order to identify pixels belonging to the breast vascular network. Two families of features were extracted: first, morphological and textural features from segmented images in order to quantify the extent and the properties of the vascular network; second, textural features from the whole breast region in order to investigate whether the nature of the disease causes statistically important changes in the texture of affected breasts. Results have indicated that: (a) the texture of vessels presents statistically significant differences (p<0.001) between normal, benign and malignant cases, (b) the texture of the whole breast region for malignant and non-malignant breasts, produced statistically significant differences (p<0.001), (c) the relative ratios of the texture between the two breasts may be used for the discrimination of non-malignant from malignant patients, and (d) an area under the receiver operating characteristic curve of 0.908 (AUC) was found when features were combined in a logistic regression prediction rule according to ROC analysis.

Discrete states of attention during active visual fixation revealed by Markovian analysis of the time series of intrusive saccades.

The frequency of intrusive saccades during maintenance of active visual fixation has been used as a measure of sustained visual attention in studies of healthy subjects as well as of neuropsychiatric patient populations. In this study, the mechanism that generates intrusive saccades during active visual fixation was investigated in a population of young healthy men performing three sustained fixation tasks (fixation to a visual target, fixation to a visual target with visual distracters, and fixation straight ahead in the dark). Markov Chain modeling of inter-saccade intervals (ISIs) was utilized. First- and second-order Markov modeling provided indications for the existence of a non-random pattern in the production of intrusive saccades. Accordingly, the system of intrusive saccade generation may operate in two "attractor" states, one in which intrusive saccades occur at short consecutive ISIs and another in which intrusive saccades occur at long consecutive ISIs. These states might correspond to two distinct states of the attention system, one of low focused - high distractibility and another of high focused - low distractibility, such as those proposed in the adaptive gain theory for the control of attention by the noradrenergic system in the brain. To the authors knowledge, this is the first time that Markov Chain modeling has been applied to the analysis of the ISIs of intrusive saccades.

Automatic identification of oculomotor behavior using pattern recognition techniques.

In this paper, a methodological scheme for identifying distinct patterns of oculomotor behavior such as saccades, microsaccades, blinks and fixations from time series of eye's angular displacement is presented. The first step of the proposed methodology involves signal detrending for artifacts removal and estimation of eye's angular velocity. Then, feature vectors from fourteen first-order statistical features are formed from each angular displacement and velocity signal using sliding, fixed-length time windows. The obtained feature vectors are used for training and testing three artificial neural network classifiers, connected in cascade. The three classifiers discriminate between blinks and non-blinks, fixations and non-fixations and saccades and microsaccades, respectively. The proposed methodology was tested on a dataset from 1392 subjects, each performing three oculomotor fixation conditions. The average overall accuracy of the three classifiers, with respect to the manual identification of eye movements by experts, was 95.9%. The proposed methodological scheme provided better results than the well-known Velocity Threshold algorithm, which was used for comparison. The findings of the present study indicate that the utilization of pattern recognition techniques in the task of identifying the various eye movements may provide accurate and robust results.

An implicit evolution scheme for active contours and surfaces based on IIR filtering.

In this work, we present an approach for implementing an implicit scheme for the numerical solution of the partial differential equation of the evolution of an active contour/surface. The proposed scheme is applicable to any variant of the traditional active contour (AC), irrespectively of the calculation of the image-based force field and it is readily applicable to explicitly parameterized active surfaces (AS). The proposed approach is formulated as an infinite impulse response (IIR) filtering of the coordinates of the contour/surface points. The poles of the filter are determined by the parameters controlling the shape of the active contour/surface. We show that the proposed IIR-based implicit evolution scheme has very low complexity. Furthermore, the proposed scheme is numerically stable, thus it allows the convergence of the AC/AS with significantly fewer iterations than the explicit evolution scheme. It also possesses the separability property along the two parameters of the AS, thus it may be applied to deformable surfaces, without the need to store and invert large sparse matrices. We implemented the proposed IIR-based implicit evolution scheme in the Vector Field Convolution (VFC) AC/AS using synthetic and clinical volumetric data. We compared the segmentation results with those of the explicit AC/AS evolution, in terms of accuracy and efficiency. Results show that the VFC AC/AS with the proposed IIR-based implicit evolution scheme achieves the same segmentation results with the explicit scheme, with considerably less computation time.

A classification system based on a new wrapper feature selection algorithm for the diagnosis of primary and secondary polycythemia.

Primary and Secondary Polycythemia are diseases of the bone marrow that affect the blood's composition and prohibit patients from becoming blood donors. Since these diseases may become fatal, their early diagnosis is important. In this paper, a classification system for the diagnosis of Primary and Secondary Polycythemia is proposed. The proposed system classifies input data into three classes; Healthy, Primary Polycythemic (PP) and Secondary Polycythemic (SP) and is implemented using two separate binary classification levels. The first level performs the Healthy/non-Healthy classification and the second level the PP/SP classification. To this end, a novel wrapper feature selection algorithm, called the LM-FM algorithm, is presented in order to maximize the classifier's performance. The algorithm is comprised of two stages that are applied sequentially: the Local Maximization (LM) stage and the Floating Maximization (FM) stage. The LM stage finds the best possible subset of a fixed predefined size, which is then used as an input for the next stage. The FM stage uses a floating size technique to search for an even better solution by varying the initially provided subset size. Then, the Support Vector Machine (SVM) classifier is used for the discrimination of the data at each classification level. The proposed classification system is compared with various well-established feature selection techniques such as the Sequential Floating Forward Selection (SFFS) and the Maximum Output Information (MOI) wrapper schemes, and with standalone classification techniques such as the Multilayer Perceptron (MLP) and SVM classifier. The proposed LM-FM feature selection algorithm combined with the SVM classifier increases the overall performance of the classification system, scoring up to 98.9% overall accuracy at the first classification level and up to 96.6% at the second classification level. Moreover, it provides excellent robustness regardless of the size of the input feature subset used.

Classification of histological images of the endometrium using texture features.

OBJECTIVE: To present a texture analysis method in order to achieve texture classification for 240 histological images of the endometrium. STUDY DESIGN: A total of 128 patients with endometrial cancer and 112 subjects with no pathological condition were imaged. For each image 190 texture features were initially extracted, derived from the wavelets, the Gabor filters, and the Law's masks, which were reduced after feature selection in only 4 features. RESULTS: The images were classified into 2 categories using artificial neural networks, and the reported classification accuracy was 98.1%. CONCLUSION: The results showed that there was a strong discrimination between histological images of cancerous and normal tissue of the endometrium, based on the proposed set of texture features.

Classification of hysteroscopical images using texture and vessel descriptors.

In recent years, hysteroscopy, used as an outpatient office procedure, in combination with endometrial biopsy, has demonstrated its great potential as the method of first choice in the diagnosis of various gynecological abnormalities including abnormal uterine bleeding (AUB) and endometrial cancer (CA). In patients suffering with AUB, the blood vessels of the endometrium are hypertrophic, whereas in the case of CA vascularization is irregular or anarchic. In this paper, a methodology for the classification of hysteroscopical images of endometrium using vessel and texture features is presented. A total of 28 patients with abnormal uterine bleeding, 10 patients with endometrial cancer and 39 subjects with no pathological condition were imaged. 16 of the patients with AUB were premenopausal and 12 postmenopausal, all with CA were postmenopausal, and all with no pathological condition were premenopausal. All images were examined for the appearance of endometrial vessels and non-vascular structures. For each image, 167 texture and vessel's features were initially extracted, which were reduced after feature selection in only 4 features. The images were classified into three categories using artificial neural networks and the reported classification accuracy was 91.2 %, while the specificity and sensitivity were 83.8 and 93.6 % respectively.

A decision support system for assisting fine needle aspiration diagnosis of thyroid malignancy.

OBJECTIVE: To assist diagnosis of thyroid malignancy, implementing a decision support system (DSS) using fine needle aspiration biopsy (FNAB) data. STUDY DESIGN: The set of 2,035 thyroid smears contained 1,886 smears of nonmalignancy (class 1) and 150 smears of malignancy (class 2) verified histologically. For each smear, 67 medical features were considered by the expert, forming 2,036 feature vectors, which were fed into a DSS for discriminating between malignant and nonmalignant smears. The DSS comprised a feature selection and classification module using a combination of three classifiers, the artificial neural network, the support vector machines, and the k-nearest neighbor, under the majority vote procedure. RESULTS: The overall classification accuracy of the DSS was 98.6%, marginally better than the FNAB (97.3%). The DSS had lower sensitivity (89.1%) and better specificity (99.4%) compared to the FNAB. Regarding the smears characterized as "suspicious" by FNAB, a significant improvement of overall accuracy was obtained by the proposed DSS system (84.6%) compared to the FNAB (50.0%). CONCLUSION: The proposed DSS provides significant improvement compared to FNAB regarding discrimination of smears characterized by an expert as "suspicious," reducing the number of patients undergoing surgical procedures.

Automatic point correspondence using an artificial immune system optimization technique for medical image registration.

In this paper, an automatic method for determining pairs of corresponding points between medical images is proposed. The method is based on the implementation of an artificial immune system (AIS). AIS is a relatively novel, population based category of algorithms, inspired by theoretical immunologic models. When used as function optimizers, AIS have the attractive property of locating the global optimum of a function as well as a large number of strong local optimum points. In this work, AIS has been applied both for the extraction of an optimal set of candidate points on the reference image and the definition of their corresponding ones on the second image. The performance of the proposed AIS algorithm is evaluated against the widely used Iterative Closest Point (ICP) algorithm in terms of the accuracy of the obtained correspondences and in terms of the accuracy of the point-based registration by the two correspondence algorithms and the Mutual Information criterion, as an intensity-based registration method. Qualitative and quantitative results involving 92 X-ray dental and 10 retinal image pairs subject to known and unknown transformations are presented. The results indicate a superior performance of the proposed AIS algorithm with respect to the ICP algorithm and the Mutual Information, in terms of both correct correspondence and registration accuracy.

Application of Kohonen network for automatic point correspondence in 2D medical images.

In this paper, a generalized application of Kohonen Network for automatic point correspondence of unimodal medical images is presented. Given a pair of two-dimensional medical images of the same anatomical region and a set of interest points in one of the images, the algorithm detects effectively the set of corresponding points in the second image, by exploiting the properties of the Kohonen self organizing maps (SOMs) and embedding them in a stochastic optimization framework. The correspondences are established by determining the parameters of local transformations that map the interest points of the first image to their corresponding points in the second image. The parameters of each transformation are computed in an iterative way, using a modification of the competitive learning, as implemented by SOMs. The proposed algorithm was tested on medical imaging data from three different modalities (CT, MR and red-free retinal images) subject to known and unknown transformations. The quantitative results in all cases exhibited sub-pixel accuracy. The algorithm also proved to work efficiently in the case of noise corrupted data. Finally, in comparison to a previously published algorithm that was also based on SOMs, as well as two widely used techniques for detection of point correspondences (template matching and iterative closest point), the proposed algorithm exhibits an improved performance in terms of accuracy and robustness.

Computer-aided diagnosis of thyroid malignancy using an artificial immune system classification algorithm.

The diagnosis of thyroid malignancy by fine needle aspiration (FNA) examination has been proven to show wide variations of sensitivity and specificity. This paper proposes the utilization of a computer-aided diagnosis system based on a supervised classification algorithm from the artificial immune systems to assist the task of thyroid malignancy diagnosis. The core of the proposed algorithm is the so-called BoxCells, which are defined as parallelepipeds in the feature space. Properly defined operators act on the BoxCells in order to convert them into individual, elementary classifiers. The proposed algorithm is applied on FNA data from 2016 subjects with verified diagnosis and has exhibited average specificity higher than 99%, 90% sensitivity, and 98.5% accuracy. Furthermore, 24% of the cases that are characterized as "suspicious" by FNA and are histologically proven nonmalignancies have been classified correctly.

An iterative point correspondence algorithm for automatic image registration: an application to dental subtraction radiography.

In this paper, an Automatic Iterative Point Correspondence (AIPC) algorithm towards image registration is presented. Given an image pair, distinctive points are extracted only in one of the images (reference image), and the corresponding points in the other image are obtained automatically by maximizing a similarity measure between regions of the two images with respect to the parameters of a local transformation. The maximization is accomplished by means of an iterative procedure, in which candidate solutions for the transformation parameters are tested at each iteration; these solutions are evaluated by the similarity measure between image regions. The detected point pairs by the application of the AIPC algorithm are then used to estimate the parameters of a global projective transformation for the registration of the image pair. The proposed AIPC algorithm was applied on 113 in vitro and in vivo dental image pairs providing improved registration accuracy against three widely used registration methods.

Detection of glaucomatous change based on vessel shape analysis.

Glaucoma, a leading cause of blindness worldwide, is a progressive optic neuropathy with characteristic structural changes in the optic nerve head and concomitant visual field defects. Ocular hypertension (i.e. elevated intraocular pressure without glaucoma) is the most important risk factor to develop glaucoma. Even though a number of variables, including various optic disc and visual field parameters, have been used in order to identify early glaucomatous damage, there is a need for computer-based methods that can detect early glaucomatous progression so that treatment to prevent further progression can be initiated. This paper is focused on the description of a system based on image processing and classification techniques for the estimation of quantitative parameters to define vessel deformation and the classification of image data into two classes: patients with ocular hypertension who develop glaucomatous damage and patients with ocular hypertension who remain stable. The proposed system consists of the retinal image preprocessing module for vessel central axis segmentation, the automatic retinal image registration module based on a novel application of self organizing maps (SOMs) to define automatic point correspondence, the retinal vessel attributes calculation module to select the vessel shape attributes and the data classification module, using an artificial neural network classifier, to perform the necessary subject classification. Implementation of the system to optic disc data from 127 subjects obtained by a fundus camera at regular intervals provided a classification rate of 87.5%, underscoring the value of the proposed system to assist in the detection of early glaucomatous change.

Artificial immune network for automatic point correspondence in medical images.

In this work, an automatic method for point-by point correspondence between medical images is presented based on the implementation of an Artificial Immune Network (AIN). AIN is a relatively novel population based algorithm, which when applied to multimodal function optimization exhibit the attractive feature of locating, the global minimum of a function, as well as a large number of strong local optimum points. In this work, AIN has been modified and applied to the problem of automatic point correspondence from pairs of images. Additionally, the proposed system is capable of altering the initially selected points on the reference image so that the population of points becomes fitter. The performance of the proposed algorithm using the AIN is evaluated against a standardized method for automatic correspondence, the template matching, in terms of the accuracy of the correspondence. Qualitative and quantitative results presented from in vitro radiographic dental images with synthetic deformations, show that the proposed algorithm outperforms the template matching for automatic point correspondence.

ESTERR-PRO: a setup verification software system using electronic portal imaging.

The purpose of the paper is to present and evaluate the performance of a new software-based registration system for patient setup verification, during radiotherapy, using electronic portal images. The estimation of setup errors, using the proposed system, can be accomplished by means of two alternate registration methods. (a) The portal image of the current fraction of the treatment is registered directly with the reference image (digitally reconstructed radiograph (DRR) or simulator image) using a modified manual technique. (b) The portal image of the current fraction of the treatment is registered with the portal image of the first fraction of the treatment (reference portal image) by applying a nearly automated technique based on self-organizing maps, whereas the reference portal has already been registered with a DRR or a simulator image. The proposed system was tested on phantom data and on data from six patients. The root mean square error (RMSE) of the setup estimates was 0.8 +/- 0.3 (mean value +/- standard deviation) for the phantom data and 0.3 +/- 0.3 for the patient data, respectively, by applying the two methodologies. Furthermore, statistical analysis by means of the Wilcoxon nonparametric signed test showed that the results that were obtained by the two methods did not differ significantly (P value >0.05).

Image registration based on lifting process: an application to digital subtraction radiography.

In this paper, a digital subtraction radiology scheme is presented based on a new method for the automatic registration of dental radiographs acquired with or without rigorous a priori standardization. The scheme is comprised of an automatic registration method and a subtraction process. The proposed registration method can be considered as an object-based registration method without imposing the prerequisite of image segmentation in order to detect the boundary of the objects of interest or the automatic detection of matching landmarks. This is achieved by augmenting the dimensionality of the problem from two-dimensional gray-level matching to three-dimensional surface matching using the process of lifting in combination with a surface-matching technique. The pseudo three-dimensional affine transformation that matches the lifted images incorporates advantageous characteristics including spatial alignment of the surfaces, anisotropic correction of brightness/contrast differences, and stable convergence of the similarity function to its optimal value. The performance of the proposed automatic registration method is assessed against a manual method based on the projective transformation. The qualitative and quantitative assessments of the experiments have shown advantageous performance of the proposed automatic registration method against the manual one. Finally, the proposed registration method has been further improved in terms of execution time by the implementation of a surface decimation process.