Intelligence against complexity: Machine learning for nonuniform temperature-field measurements through laser absorption

The effect of spatial nonuniformity of the temperature distribution was examined on the capability of machine-learning algorithms to provide accurate temperature prediction based on Laser Absorption Spectroscopy. First, sixteen machine learning models were trained as surrogate models of conventional physical methods to measure temperature from uniform temperature distributions (uniform-profile spectra). The best three of them, Gaussian Process Regression (GPR), VGG13, and Boosted Random Forest (BRF) were shown to work excellently on uniform profiles but their performance degraded tremendously on nonuniform-profile spectra. This indicated that directly using uniform-profile-targeted methods to nonuniform profiles was improper. However, after retraining models on nonuniform-profile data, the models of GPR and VGG13, which utilized all features of the spectra, not only showed good accuracy and sensitivity to spectral twins, but also showed excellent generalization performance on spectra of increased nonuniformity, which demonstrated that the negative effects of nonuniformity on temperature measurement could be overcome. In contrast, BRF, which utilized partial features, did not have good generalization performance, which implied the nonuniformity level had impact on regional features of spectra. By reducing the data dimensionality through T-SNE and LDA, the visualizations of the data in two-dimensional feature spaces demonstrated that two datasets of substantially different levels of non-uniformity shared very closely similar distributions in terms of both spectral appearance and spectrum-temperature mapping. Notably, datasets from uniform and nonuniform temperature distributions clustered in two different areas of the 2D spaces of the t-SNE and LDA features with very few samples overlapping.

Local-Sensitive Connectivity Filter (LS-CF): A Post-Processing Unsupervised Improvement of the Frangi, Hessian and Vesselness Filters for Multimodal Vessel Segmentation

A retinal vessel analysis is a procedure that can be used as an assessment of risks to the eye. This work proposes an unsupervised multimodal approach that improves the response of the Frangi filter, enabling automatic vessel segmentation. We propose a filter that computes pixel-level vessel continuity while introducing a local tolerance heuristic to fill in vessel discontinuities produced by the Frangi response. This proposal, called the local-sensitive connectivity filter (LS-CF), is compared against a naive connectivity filter to the baseline thresholded Frangi filter response and to the naive connectivity filter response in combination with the morphological closing and to the current approaches in the literature. The proposal was able to achieve competitive results in a variety of multimodal datasets. It was robust enough to outperform all the state-of-the-art approaches in the literature for the OSIRIX angiographic dataset in terms of accuracy and 4 out of 5 works in the case of the IOSTAR dataset while also outperforming several works in the case of the DRIVE and STARE datasets and 6 out of 10 in the CHASE-DB dataset. For the CHASE-DB, it also outperformed all the state-of-the-art unsupervised methods.

Analysing the impact of global demographic characteristics over the COVID-19 spread using class rule mining and pattern matching

Since the coronavirus disease (COVID-19) outbreak in December 2019, studies have been addressing diverse aspects in relation to COVID-19 and Variant of Concern 202012/01 (VOC 202012/01) such as potential symptoms and predictive tools. However, limited work has been performed towards the modelling of complex associations between the combined demographic attributes and varying nature of the COVID-19 infections across the globe. This study presents an intelligent approach to investigate the multi-dimensional associations between demographic attributes and COVID-19 global variations. We gather multiple demographic attributes and COVID-19 infection data (by 8 January 2021) from reliable sources, which are then processed by intelligent algorithms to identify the significant associations and patterns within the data. Statistical results and experts' reports indicate strong associations between COVID-19 severity levels across the globe and certain demographic attributes, e.g. female smokers, when combined together with other attributes. The outcomes will aid the understanding of the dynamics of disease spread and its progression, which in turn may support policy makers, medical specialists and society, in better understanding and effective management of the disease.

ELEMENT: Multi-Modal Retinal Vessel Segmentation Based on a Coupled Region Growing and Machine Learning Approach

Vascular structures in the retina contain important information for the detection and analysis of ocular diseases, including age-related macular degeneration, diabetic retinopathy and glaucoma. Commonly used modalities in diagnosis of these diseases are fundus photography, scanning laser ophthalmoscope (SLO) and fluorescein angiography (FA). Typically, retinal vessel segmentation is carried out either manually or interactively, which makes it time consuming and prone to human errors. In this research, we propose a new multi-modal framework for vessel segmentation called ELEMENT (vEsseL sEgmentation using Machine lEarning and coNnecTivity). This framework consists of feature extraction and pixel-based classification using region growing and machine learning. The proposed features capture complementary evidence based on grey level and vessel connectivity properties. The latter information is seamlessly propagated through the pixels at the classification phase. ELEMENT reduces inconsistencies and speeds up the segmentation throughput. We analyze and compare the performance of the proposed approach against state-of-the-art vessel segmentation algorithms in three major groups of experiments, for each of the ocular modalities. Our method produced higher overall performance, with an overall accuracy of 97.40%, compared to 25 of the 26 state-of-the-art approaches, including six works based on deep learning, evaluated on the widely known DRIVE fundus image dataset. In the case of the STARE, CHASE-DB, VAMPIRE FA, IOSTAR SLO and RC-SLO datasets, the proposed framework outperformed all of the state-of-the-art methods with accuracies of 98.27%, 97.78%, 98.34%, 98.04% and 98.35%, respectively.

Application of Convolutional Neural Networks for Automated Ulcer Detection in Wireless Capsule Endoscopy Images

Detection of abnormalities in wireless capsule endoscopy (WCE) images is a challenging task. Typically, these images suffer from low contrast, complex background, variations in lesion shape and color, which affect the accuracy of their segmentation and subsequent classification. This research proposes an automated system for detection and classification of ulcers in WCE images, based on state-of-the-art deep learning networks. Deep learning techniques, and in particular, convolutional neural networks (CNNs), have recently become popular in the analysis and recognition of medical images. The medical image datasets used in this study were obtained from WCE video frames. In this work, two milestone CNN architectures, namely the AlexNet and the GoogLeNet are extensively evaluated in object classification into ulcer or non-ulcer. Furthermore, we examine and analyze the images identified as containing ulcer objects to evaluate the efficiency of the utilized CNNs. Extensive experiments show that CNNs deliver superior performance, surpassing traditional machine learning methods by large margins, which supports their effectiveness as automated diagnosis tools.

A hybrid energy model for region based curve evolution - Application to CTA coronary segmentation

BACKGROUND AND OBJECTIVE

State-of-the-art medical imaging techniques have enabled non-invasive imaging of the internal organs. However, high volumes of imaging data make manual interpretation and delineation of abnormalities cumbersome for clinicians. These challenges have driven intensive research into efficient medical image segmentation. In this work, we propose a hybrid region-based energy formulation for effective segmentation in computed tomography angiography (CTA) imagery.

METHODS

The proposed hybrid energy couples an intensity-based local term with an efficient discontinuity-based global model of the image for optimal segmentation. The segmentation is achieved using a level set formulation due to the computational robustness. After validating the statistical significance of the hybrid energy, we applied the proposed model to solve an important clinical problem of 3D coronary segmentation. An improved seed detection method is used to initialize the level set evolution. Moreover, we employed an auto-correction feature that captures the emerging peripheries during the curve evolution for completeness of the coronary tree.

RESULTS

We evaluated the segmentation accuracy of the proposed energy model against the existing techniques in two stages. Qualitative and quantitative results demonstrate the effectiveness of the proposed framework with a consistent mean sensitivity and specificity measures of 80% across the CTA data. Moreover, a high degree of agreement with respect to the inter-observer differences justifies the generalization of the proposed method.

CONCLUSIONS

The proposed method is effective to segment the coronary tree from the CTA volume based on hybrid image based energy, which can improve the clinicians ability to detect arterial abnormalities.

Automated recognition of the pericardium contour on processed CT images using genetic algorithms

This work proposes the use of Genetic Algorithms (GA) in tracing and recognizing the pericardium contour of the human heart using Computed Tomography (CT) images. We assume that each slice of the pericardium can be modelled by an ellipse, the parameters of which need to be optimally determined. An optimal ellipse would be one that closely follows the pericardium contour and, consequently, separates appropriately the epicardial and mediastinal fats of the human heart. Tracing and automatically identifying the pericardium contour aids in medical diagnosis. Usually, this process is done manually or not done at all due to the effort required. Besides, detecting the pericardium may improve previously proposed automated methodologies that separate the two types of fat associated to the human heart. Quantification of these fats provides important health risk marker information, as they are associated with the development of certain cardiovascular pathologies. Finally, we conclude that GA offers satisfiable solutions in a feasible amount of processing time.

Machine learning in the prediction of cardiac epicardial and mediastinal fat volumes

We propose a methodology to predict the cardiac epicardial and mediastinal fat volumes in computed tomography images using regression algorithms. The obtained results indicate that it is feasible to predict these fats with a high degree of correlation, thus alleviating the requirement for manual or automatic segmentation of both fat volumes. Instead, segmenting just one of them suffices, while the volume of the other may be predicted fairly precisely. The correlation coefficient obtained by the Rotation Forest algorithm using MLP Regressor for predicting the mediastinal fat based on the epicardial fat was 0.9876, with a relative absolute error of 14.4% and a root relative squared error of 15.7%. The best correlation coefficient obtained in the prediction of the epicardial fat based on the mediastinal was 0.9683 with a relative absolute error of 19.6% and a relative squared error of 24.9%. Moreover, we analysed the feasibility of using linear regressors, which provide an intuitive interpretation of the underlying approximations. In this case, the obtained correlation coefficient was 0.9534 for predicting the mediastinal fat based on the epicardial, with a relative absolute error of 31.6% and a root relative squared error of 30.1%. On the prediction of the epicardial fat based on the mediastinal fat, the correlation coefficient was 0.8531, with a relative absolute error of 50.43% and a root relative squared error of 52.06%. In summary, it is possible to speed up general medical analyses and some segmentation and quantification methods that are currently employed in the state-of-the-art by using this prediction approach, which consequently reduces costs and therefore enables preventive treatments that may lead to a reduction of health problems.

Hessian-LoG filtering for enhancement and detection of photoreceptor cells in adaptive optics retinal images

Automated analysis of retinal images plays a vital role in the examination, diagnosis, and prognosis of healthy and pathological retinas. Retinal disorders and the associated visual loss can be interpreted via quantitative correlations, based on measurements of photoreceptor loss. Therefore, it is important to develop reliable tools for identification of photoreceptor cells. In this paper, an automated algorithm is proposed, based on the use of the Hessian-Laplacian of Gaussian filter, which allows enhancement and detection of photoreceptor cells. The performance of the proposed technique is evaluated on both synthetic and high-resolution retinal images, in terms of packing density. The results on the synthetic data were compared against ground truth as well as cone counts obtained by the Li and Roorda algorithm. For the synthetic datasets, our method showed an average detection accuracy of 98.8%, compared to 93.9% for the Li and Roorda approach. The packing density estimates calculated on the retinal datasets were validated against manual counts and the results obtained by a proprietary software from Imagine Eyes and the Li and Roorda algorithm. Among the tested methods, the proposed approach showed the closest agreement with manual counting.

Automatic segmentation of coronary arteries in CT imaging in the presence of kissing vessel artifacts

In this paper, we present a novel two-step algorithm for segmentation of coronary arteries in computed tomography images based on the framework of active contours. In the proposed method, both global and local intensity information is utilized in the energy calculation. The global term is defined as a normalized cumulative distribution function, which contributes to the overall active contour energy in an adaptive fashion based on image histograms, to deform the active contour away from local stationary points. Possible outliers, such as kissing vessel artifacts, are removed in the postprocessing stage by a slice-by-slice correction scheme based on multiregion competition, where both arteries and kissing vessels are identified and tracked through the slices. The efficiency and the accuracy of the proposed technique are demonstrated on both synthetic and real datasets. The results on clinical datasets show that the method is able to extract the major branches of arteries with an average distance of 0.73 voxels to the manually delineated ground truth data. In the presence of kissing vessel artifacts, the outer surface of the entire coronary tree, extracted by the proposed algorithm, is smooth and contains fewer erroneous regions, originating in kissing vessel artifacts, as compared to the initial segmentation.

3-D quantitative vascular shape analysis for arterial bifurcations via dynamic tube fitting

Reliable and reproducible estimation of vessel centerlines and reference surfaces is an important step for the assessment of luminal lesions. Conventional methods are commonly developed for quantitative analysis of the "straight" vessel segments and have limitations in defining the precise location of the centerline and the reference lumen surface for both the main vessel and the side branches in the vicinity of bifurcations. To address this, we propose the estimation of the centerline and the reference surface through the registration of an elliptical cross-sectional tube to the desired constituent vessel in each major bifurcation of the arterial tree. The proposed method works directly on the mesh domain, thus alleviating the need for image upsampling, usually required in conventional volume domain approaches. We demonstrate the efficiency and accuracy of the method on both synthetic images and coronary CT angiograms. Experimental results show that the new method is capable of estimating vessel centerlines and reference surfaces with a high degree of agreement to those obtained through manual delineation. The centerline errors are reduced by an average of 62.3% in the regions of the bifurcations, when compared to the results of the initial solution obtained through the use of mesh contraction method.

On sparse forward solutions in non-stationary domains for the EIT imaging problem

In the forward EIT-problem numerical solutions of an elliptic partial differential equation are required. Given the arbitrary geometries encountered, the Finite Element Method (FEM) is, naturally, the method of choice. Nowadays, in EIT applications, there is an increasing demand for finer Finite Element mesh models. This in turn results to a soaring number of degrees of freedom and an excessive number of unknowns. As such, only piece-wise linear basis functions can practically be employed to maintain inexpensive computations. In addition, domain reduction and/or compression schemes are often sought to further counteract for the growing number of unknowns. In this paper, we replace the piece-wise linear with wavelet basis functions (coupled with the domain embedding method) to enable sparse approximations of the forward computations. Given that the forward solutions are repeatedly, if not extensively, utilised during the image reconstruction process, considerable computational savings can be recorded whilst maintaining O(N) forward problem complexity. We verify with numerical results that, in practice, less than 5% of the involved coefficients are actually required for computations and, hence, needs to be stored. We finalise this work by addressing the impact to the inverse problem. It is worth underlining that the proposed scheme is independent of the actual family of wavelet basis functions of compact support.

A fisher information matrix interpretation of the NOSER algorithm in electrical impedance tomography

In this paper, we employ the concept of the Fisher information matrix (FIM) to reformulate and improve on the "Newton's One-Step Error Reconstructor" (NOSER) algorithm. FIM is a systematic approach for incorporating statistical properties of noise, modeling errors and multi-frequency data. The method is discussed in a maximum likelihood estimator (MLE) setting. The ill-posedness of the inverse problem is mitigated by means of a nonlinear regularization strategy. It is shown that the overall approach reduces to the maximum a posteriori estimator (MAP) with the prior (conductivity vector) described by a multivariate normal distribution. The covariance matrix of the prior is a diagonal matrix and is computed directly from the Fisher information matrix. An eigenvalue analysis is presented, revealing the advantages of using this prior to a Gaussian smoothness prior (Laplace). Reconstructions are shown using measured data obtained from a shallow breathing of an adult human subject. The reconstructions show that the FIM approach clearly improves on the original NOSER algorithm.

Towards non-invasive EIT imaging of domains with deformable boundaries

We investigate on the use of the Domain Embedding Method (DEM) for the forward modelling in EIT. This approach is suitably configured to overcome the model meshing bottleneck since it does not require that the mesh on the domain is adapted to the boundary surface. This is of crucial importance for, e.g., clinical applications of EIT, as it avoids tedious and time-consuming (re-)meshing procedures. The suggested DEM approach can accommodate arbitrary yet Lipschitz smooth boundary surfaces and is not limited to polygonal domains. For the discretisation purposes, we employ B-splines as they allow for arbitrary accuracy by raising the polynomial degree and are easy to implement due to their inherent piecewise polynomial structure. Numerical experiments confirm that a B-spline discretization yields, similarly to conventional Finite Difference discretizations, increasing condition numbers of the system matrix with respect to the discretisation levels. Fortunately, multiresolution ideas based on B-splines allow for optimal wavelet preconditioning.

Generalized regression neural networks with multiple-bandwidth sharing and hybrid optimization

This paper proposes a novel algorithm for function approximation that extends the standard generalized regression neural network. Instead of a single bandwidth for all the kernels, we employ a multiple-bandwidth configuration. However, unlike previous works that use clustering of the training data for the reduction of the number of bandwidths, we propose a distinct scheme that manages a dramatic bandwidth reduction while preserving the required model complexity. In this scheme, the algorithm partitions the training patterns to groups, where all patterns within each group share the same bandwidth. Grouping relies on the analysis of the local nearest neighbor distance information around the patterns and the principal component analysis with fuzzy clustering. Furthermore, we use a hybrid optimization procedure combining a very efficient variant of the particle swarm optimizer and a quasi-Newton method for global optimization and locally optimal fine-tuning of the network bandwidths. Training is based on the minimization of a flexible adaptation of the leave-one-out validation error that enhances the network generalization. We test the proposed algorithm with real and synthetic datasets, and results show that it exhibits competitive regression performance compared to other techniques.

Density-driven generalized regression neural networks (DD-GRNN) for function approximation

This paper proposes a new nonparametric regression method, based on the combination of generalized regression neural networks (GRNNs), density-dependent multiple kernel bandwidths, and regularization. The presented model is generic and substitutes the very large number of bandwidths with a much smaller number of trainable weights that control the regression model. It depends on sets of extracted data density features which reflect the density properties and distribution irregularities of the training data sets. We provide an efficient initialization scheme and a second-order algorithm to train the model, as well as an overfitting control mechanism based on Bayesian regularization. Numerical results show that the proposed network manages to reduce significantly the computational demands of having individual bandwidths, while at the same time, provides competitive function approximation accuracy in relation to existing methods.

Near-infrared spectroscopic measurements of blood analytes using multi-layer perceptron neural networks

Near-infrared (NIR) spectroscopy is being applied to the solution of problems in many areas of biomedical and pharmaceutical research. In this paper we investigate the use of NIR spectroscopy as an analytical tool to quantify concentrations of urea, creatinine, glucose and oxyhemoglobin (HbO2). Measurements have been made in vitro with a portable spectrometer developed in our labs that consists of a two beam interferometer operating in the range of 800-2300 nm. For the data analysis a pattern recognition philosophy was used with a preprocessing stage and a multi-layer perceptron (MLP) neural network for the measurement stage. Results show that the interferogram signatures of the above compounds are sufficiently strong in that spectral range. Measurements of three different concentrations were possible with mean squared error (MSE) of the order of 10(-6).