Semi-supervised COVID-19 volumetric pulmonary lesion estimation on CT images using probabilistic active contour and CNN segmentation.

Purpose: A semi-supervised two-step methodology is proposed to obtain a volumetric estimation of COVID-19-related lesions on Computed Tomography (CT) images. Methods: First, damaged tissue was segmented from CT images using a probabilistic active contours approach. Second, lung parenchyma was extracted using a previously trained U-Net. Finally, volumetric estimation of COVID-19 lesions was calculated considering the lung parenchyma masks.Our approach was validated using a publicly available dataset containing 20 CT COVID-19 images previously labeled and manually segmented. Then, it was applied to 295 COVID-19 patients CT scans admitted to an intensive care unit. We compared the lesion estimation between deceased and survived patients for high and low-resolution images. Results: A comparable median Dice similarity coefficient of 0.66 for the 20 validation images was achieved. For the 295 images dataset, results show a significant difference in lesion percentages between deceased and survived patients, with a p-value of 9.1 × 10-4 in low-resolution and 5.1 × 10-5 in high-resolution images. Furthermore, the difference in lesion percentages between high and low-resolution images was 10 % on average. Conclusion: The proposed approach could help estimate the lesion size caused by COVID-19 in CT images and may be considered an alternative to getting a volumetric segmentation for this novel disease without the requirement of large amounts of COVID-19 labeled data to train an artificial intelligence algorithm. The low variation between the estimated percentage of lesions in high and low-resolution CT images suggests that the proposed approach is robust, and it may provide valuable information to differentiate between survived and deceased patients.

Assessment of Singularities in the EEG During A-Phases of Sleep Based on Wavelet Decomposition.

Electroencephalography (EEG) signals convey information related to different processes that take place in the brain. From the EEG fluctuations during sleep, it is possible to establish the sleep stages and identify short events, commonly related to a specific physiological process or pathology. Some of these short events (called A-phases) present an organization and build up the concept of the Cyclic Alternating Pattern (CAP) phenomenon. In general, the A-phases abruptly modify the EEG fluctuations, and a singular behavior could occur. With the aim to quantify the abrupt changes during A-phases, in this work the wavelet analysis is considered to compute Hölder exponents, which measure the singularity strength. We considered time windows of 2s outside and 5s inside A-phases onset (or offset). A total number of 5121 A-phases from 9 healthy participants and 10 patients with periodic leg movements were analyzed. Within an A-phase the Hölder numerical value tends to be 0.6, which implies a less abrupt singularity. Whereas outside of A-phases, it is observed that the Hölder value is approximately equal to 0.3, which implies stronger singularities, i.e., a more evident discontinuity in the signal behavior. In addition, it seems that the number of singularities increases inside of A-phases. The numerical results suggest that the EEG naturally conveys singularities modified by the A-phase occurrence, and this information could help to conceptualize the CAP phenomenon from a new perspective based on the sharpness of the EEG instead of the oscillatory way.

COVID-19 Volumetric Pulmonary Lesion Estimation on CT Images using a U-NET and Probabilistic Active Contour Segmentation.

A two-step method for obtaining a volumetric estimation of COVID-19 related lesion from CT images is proposed. The first step consists in applying a U-NET convolutional neural network to provide a segmentation of the lung-parenchyma. This architecture is trained and validated using the Thoracic Volume and Pleural Effusion Segmentations in Diseased Lungs for Benchmarking Chest CT Processing Pipelines (PleThora) dataset, which is publicly available. The second step consists in obtaining the volumetric lesion estimation using an automatic algorithm based on a probabilistic active contour (PACO) region delimitation approach. Our pipeline successfully segmented COVID-19 related lesions in CT images, with exception of some mislabeled regions including lung airways and vasculature. Our workflow was applied to images in a cohort of 50 patients.

Probabilistic Multiple Sclerosis Lesion Detection using Superpixels and Markov Random Fields / Detección Probabilística de Lesiones de Esclerosis Múltiple usando Superpixeles y Campos Aleatorios de Markov

Abstract Multiple Sclerosis (MS) is the most common neurodegenerative disease among young adults. Diagnosis and monitoring of MS is performed with T2-weighted or T2 FLAIR magnetic resonance imaging, where MS lesions appear as hyperintense spots in the white matter. In recent years, multiple algorithms have been proposed to detect these lesions with varying success rates, which greatly depend on the amount of a priori information required by each algorithm, such as the use of an atlas or the involvement of an expert to guide the segmentation process. In this work, a fully automatic method that does not rely on a priori anatomical information is proposed and evaluated. The proposed algorithm is based on an over-segmentation in superpixels and their classification by means of Gauss-Markov Measure Fields (GMMF). The main advantage of the over-segmentation is that it preserves the borders between tissues, while the GMMF classifier is robust to noise and computationally efficient. The proposed segmentation is then applied in two stages: first to segment the brain region and then to detect hyperintense spots within the brain. The proposed method is evaluated with synthetic images from BrainWeb, as well as real images from MS patients. The proposed method produces competitive results with respect to other algorithms in the state of the art, without requiring user assistance nor anatomical prior information.

Resumen La Esclerosis Múltiple (MS) es una de las enfermedades neurodegenerativas más comunes en adultos jóvenes. El diagnóstico y su monitoreo se realiza generalmente mediante imágenes de resonancia magnética T2 o T2 FLAIR, donde se observan regiones hiperintensas relacionadas a lesiones cerebrales causadas por la MS. En años recientes, múltiples algoritmos han sido propuestos para detectar estas lesiones con diferentes tasas de éxito las cuales dependen en gran medida de la cantidad de información a priori que requiere cada algoritmo, como el uso de un atlas o el involucramiento de un experto que guíe el proceso de segmentación. En este trabajo, se propone un método automático independiente de información anatómica. El algoritmo propuesto está basado en una sobresegmentación en superpixeles y su clasificación mediante un proceso de Campos Aleatorios de Markov de Medidas Gaussianas (GMMF). La principal ventaja de la sobresegmentación es que preserva bordes entre tejidos, además que tiene un costo reducido en tiempo de ejecución, mientras que el clasificador GMMF es robusto a ruido y computacionalmente eficiente. La segmentación propuesta es aplicada en dos etapas: primero para segmentar el cerebro y después para detectar las lesiones en él. El método propuesto es evaluado usando imágenes sintéticas de BrainWeb, así como también imágenes reales de pacientes con MS. Con respecto a los resultados, el método propuesto muestra un desempeño competitivo respecto a otros métodos en el estado del arte, tomando en cuenta que éste no requiere de asistencia o información a priori.

A-phase classification using convolutional neural networks.

A series of short events, called A-phases, can be observed in the human electroencephalogram (EEG) during Non-Rapid Eye Movement (NREM) sleep. These events can be classified in three groups (A1, A2, and A3) according to their spectral contents, and are thought to play a role in the transitions between the different sleep stages. A-phase detection and classification is usually performed manually by a trained expert, but it is a tedious and time-consuming task. In the past two decades, various researchers have designed algorithms to automatically detect and classify the A-phases with varying degrees of success, but the problem remains open. In this paper, a different approach is proposed: instead of attempting to design a general classifier for all subjects, we propose to train ad-hoc classifiers for each subject using as little data as possible, in order to drastically reduce the amount of time required from the expert. The proposed classifiers are based on deep convolutional neural networks using the log-spectrogram of the EEG signal as input data. Results are encouraging, achieving average accuracies of 80.31% when discriminating between A-phases and non A-phases, and 71.87% when classifying among A-phase sub-types, with only 25% of the total A-phases used for training. When additional expert-validated data is considered, the sub-type classification accuracy increases to 78.92%. These results show that a semi-automatic annotation system with assistance from an expert could provide a better alternative to fully automatic classifiers. Graphical abstract A/N Deep Learning Classifier.

A new Probabilistic Active Contour region-based method for multiclass medical image segmentation.

In medical imaging, the availability of robust and accurate automatic segmentation methods is very important for a user-independent and time-saving delineation of regions of interest. In this work, we present a new variational formulation for multiclass image segmentation based on active contours and probability density functions demonstrating that the method is fast, accurate, and effective for MRI brain image segmentation. We define an energy function assuming that the regions to segment are independent. The first term of this function measures how much the pixels belong to each class and forces the regions to be disjoint. In order for this term to be outlier-resistant, probability density functions were used allowing to define the structures to be segmented. The second one is the classical regularization term which constrains the border length of each region removing inhomogeneities and noise. Experiments with synthetic and real images showed that this approach is robust to noise and presents an accuracy comparable to other classical segmentation approaches (in average DICE coefficient over 90% and ASD below one pixel), with further advantages related to segmentation speed. Graphical Abstract.

Analysis of A-phase transitions during the cyclic alternating pattern under normal sleep.

An analysis of the EEG signal during the B-phase and A-phases transitions of the cyclic alternating pattern (CAP) during sleep is presented. CAP is a sleep phenomenon composed by consecutive sequences of A-phases (each A-phase could belong to a possible group A1, A2 or A3) observed during the non-REM sleep. Each A-phase is separated by a B-phase which has the basal frequency of the EEG during a specific sleep stage. The patterns formed by these sequences reflect the sleep instability and consequently help to understand the sleep process. Ten recordings from healthy good sleepers were included in this study. The current study investigates complexity, statistical and frequency signal properties of electroencephalography (EEG) recordings at the transitions: B-phase--A-phase. In addition, classification between the onset-offset of the A-phases and B-phase was carried out with a kNN classifier. The results showed that EEG signal presents significant differences (p < 0.05) between A-phases and B-phase for the standard deviation, energy, sample entropy, Tsallis entropy and frequency band indices. The A-phase onset showed values of energy three times higher than B-phase at all the sleep stages. The statistical analysis of variance shows that more than 80% of the A-phase onset and offset is significantly different from the B-phase. The classification performance between onset or offset of A-phases and background showed classification values over 80% for specificity and accuracy and 70% for sensitivity. Only during the A3-phase, the classification was lower. The results suggest that neural assembles that generate the basal EEG oscillations during sleep present an over-imposed coordination for a few seconds due to the A-phases. The main characteristics for automatic separation between the onset-offset A-phase and the B-phase are the energy at the different frequency bands.

Blind deconvolution estimation of fluorescence measurements through quadratic programming.

Time-deconvolution of the instrument response from fluorescence lifetime imaging microscopy (FLIM) data is usually necessary for accurate fluorescence lifetime estimation. In many applications, however, the instrument response is not available. In such cases, a blind deconvolution approach is required. An iterative methodology is proposed to address the blind deconvolution problem departing from a dataset of FLIM measurements. A linear combination of a base conformed by Laguerre functions models the fluorescence impulse response of the sample at each spatial point in our formulation. Our blind deconvolution estimation (BDE) algorithm is formulated as a quadratic approximation problem, where the decision variables are the samples of the instrument response and the scaling coefficients of the basis functions. In the approximation cost function, there is a bilinear dependence on the decision variables. Hence, due to the nonlinear nature of the estimation process, an alternating least-squares scheme iteratively solves the approximation problem. Our proposal searches for the samples of the instrument response with a global perspective, and the scaling coefficients of the basis functions locally at each spatial point. First, the iterative methodology relies on a least-squares solution for the instrument response, and quadratic programming for the scaling coefficients applied just to a subset of the measured fluorescence decays to initially estimate the instrument response to speed up the convergence. After convergence, the final stage computes the fluorescence impulse response at all spatial points. A comprehensive validation stage considers synthetic and experimental FLIM datasets of ex vivo atherosclerotic plaques and human breast cancer cell samples that highlight the advantages of the proposed BDE algorithm under different noise and initial conditions in the iterative scheme and parameters of the proposal.

An innovative multimodal/multispectral image registration method for medical images based on the Expectation-Maximization algorithm.

In this paper, we present a methodology for multimodal/ multispectral image registration of medical images. This approach is formulated by using the Expectation-Maximization (EM) methodology, such that we estimate the parameters of a geometric transformation that aligns multimodal/multispectral images. In this framework, the hidden random variables are associated to the intensity relations between the studied images, which allow to compare multispectral intensity values between images of different modalities. The methodology is basically composed by an iterative two-step procedure, where at each step, a new estimation of the joint conditional multispectral intensity distribution and the geometric transformation is computed. The proposed algorithm was tested with different kinds of medical images, and the obtained results show that the proposed methodology can be used to efficiently align multimodal/multispectral medical images.

Estimation of the number of fluorescent end-members for quantitative analysis of multispectral FLIM data.

Multispectral fluorescence lifetime imaging (m-FLIM) can potentially allow identifying the endogenous fluorophores present in biological tissue. Quantitative description of such data requires estimating the number of components in the sample, their characteristic fluorescent decays, and their relative contributions or abundances. Unfortunately, this inverse problem usually requires prior knowledge about the data, which is seldom available in biomedical applications. This work presents a new methodology to estimate the number of potential endogenous fluorophores present in biological tissue samples from time-domain m-FLIM data. Furthermore, a completely blind linear unmixing algorithm is proposed. The method was validated using both synthetic and experimental m-FLIM data. The experimental m-FLIM data include in-vivo measurements from healthy and cancerous hamster cheek-pouch epithelial tissue, and ex-vivo measurements from human coronary atherosclerotic plaques. The analysis of m-FLIM data from in-vivo hamster oral mucosa identified healthy from precancerous lesions, based on the relative concentration of their characteristic fluorophores. The algorithm also provided a better description of atherosclerotic plaques in term of their endogenous fluorophores. These results demonstrate the potential of this methodology to provide quantitative description of tissue biochemical composition.

Blind end-member and abundance extraction for multispectral fluorescence lifetime imaging microscopy data.

This paper proposes a new blind end-member and abundance extraction (BEAE) method for multispectral fluorescence lifetime imaging microscopy (m-FLIM) data. The chemometrical analysis relies on an iterative estimation of the fluorescence decay end-members and their abundances. The proposed method is based on a linear mixture model with positivity and sum-to-one restrictions on the abundances and end-members to compensate for signature variability. The synthesis procedure depends on a quadratic optimization problem, which is solved by an alternating least-squares structure over convex sets. The BEAE strategy only assumes that the number of components in the analyzed sample is known a spriori. The proposed method is first validated by using synthetic m-FLIM datasets at 15, 20, and 25 dB signal-to-noise ratios. The samples simulate the mixed response of tissue containing multiple fluorescent intensity decays. Furthermore, the results were also validated with six m-FLIM datasets from fresh postmortem human coronary atherosclerotic plaques. A quantitative evaluation of the BEAE was made against two popular techniques: minimum volume constrained nonnegative matrix factorization (MVC-NMF) and multivariate curve resolution-alternating least-squares (MCR-ALS). Our proposed method (BEAE) was able to provide more accurate estimations of the end-members: 0.32% minimum relative error and 13.82% worst-case scenario, despite different initial conditions in the iterative optimization procedure and noise effect. Meanwhile, MVC-NMF and MCR-ALS presented more variability in estimating the end-members: 0.35% and 0.34% for minimum errors and 15.31% and 13.25% in the worst-case scenarios, respectively. This tendency was also maintained for the abundances, where BEAE obtained 0.05 as the minimum absolute error and 0.12 in the worst-case scenario; MCR-ALS and MVC-NMF achieved 0.04 and 0.06 for the minimum absolute errors, and 0.15 and 0.17 under the worst-case conditions, respectively. In addition, the average computation time was evaluated for the synthetic datasets, where MVC-NMF achieved the fastest time, followed by BEAE and finally MCR-ALS. Consequently, BEAE improved MVC-NMF in convergence to a local optimal solution and robustness against signal variability, and it is roughly 3.6 time faster than MCR-ALS.

A fully constrained optimization method for time-resolved multispectral fluorescence lifetime imaging microscopy data unmixing.

This paper presents a new unmixing methodology of multispectral fluorescence lifetime imaging microscopy (m-FLIM) data, in which the spectrum is defined as the combination of time-domain fluorescence decays at multiple emission wavelengths. The method is based on a quadratic constrained optimization (CO) algorithm that provides a closed-form solution under equality and inequality restrictions. In this paper, it is assumed that the time-resolved fluorescence spectrum profiles of the constituent components are linearly independent and known a priori. For comparison purposes, the standard least squares (LS) solution and two constrained versions nonnegativity constrained least squares (NCLS) and fully constrained least squares (FCLS) (Heinz and Chang, 2001) are also tested. Their performance was evaluated by using synthetic simulations, as well as imaged samples from fluorescent dyes and ex vivo tissue. In all the synthetic evaluations, the CO obtained the best accuracy in the estimations of the proportional contributions. CO could achieve an improvement ranging between 41% and 59% in the relative error compared to LS, NCLS, and FCLS at different signal-to-noise ratios. A liquid mixture of fluorescent dyes was also prepared and imaged in order to provide a controlled scenario with real data, where CO and FCLS obtained the best performance. The CO and FCLS were also tested with 20 ex vivo samples of human coronary arteries, where the expected concentrations are qualitatively known. A certainty measure was employed to assess the confidence in the estimations made by each algorithm. The experiments confirmed a better performance of CO, since this method is optimal with respect to equality and inequality restrictions in the linear unmixing formulation. Thus, the evaluation showed that CO achieves an accurate characterization of the samples. Furthermore, CO is a computational efficient alternative to estimate the abundance of components in m-FLIM data, since a global optimal solution is always guaranteed in a closed form.

Approximate string matching using phase correlation.

A novel method for approximate string matching with applications to bioinformatics is presented in this paper. Unlike most methods in the literature, the proposed method does not depend on the computation of the edit distance between two sequences, but uses instead a similarity index obtained by applying the phase correlation method. The resulting algorithm provides a finer control over the false positive rate, allowing users to pick out relevant matchings in less time, and can be applied for both offline and online processing.

Mesh-based approach for the 3D analysis of anatomical structures of interest in radiotherapy.

In this paper a method based on mesh surfaces approximations for the 3D analysis of anatomical structures in Radiotherapy (RT) is presented. Parotid glands meshes constructed from Megavoltage CT (MVCT) images were studied in terms of volume, distance between center of mass (distCOM) of the right and left parotids, dice similarity coefficient (DICE), maximum distance between meshes (DMax) and the average symmetric distance (ASD). A comparison with the standard binary images approach was performed. While absence of significant differences in terms of volume, DistCOM and DICE indices suggests that both approaches are comparable, the fact that the ASD showed significant difference (p=0.002) and the DMax was almost significant (p=0.053) suggests that the mesh approach should be adopted to provide accurate comparison between 3D anatomical structures of interest in RT.

Automatic sleep staging based on ballistocardiographic signals recorded through bed sensors.

This study presents different methods for automatic sleep classification based on heart rate variability (HRV), respiration and movement signals recorded through bed sensors. Two methods for feature extraction have been implemented: time variant-autoregressive model (TVAM) and wavelet discrete transform (WDT); the obtained features are fed into two classifiers: Quadratic (QD) and Linear (LD) discriminant for staging sleep in REM, nonREM and WAKE periods. The performances of all the possible combinations of feature extractors and classifiers are compared in terms of accuracy and kappa index, using clinical polysomographyc evaluation as golden standard. 17 recordings from healthy subjects, including also polisomnography, were used to train and test the algorithms. When automatic classification is compared. QD-TVAM algorithm achieved a total accuracy of 76.81 ± 7.51 % and kappa index of 0.55 ± 0.10, while LD-WDT achieved a total accuracy of 79 ± 10% and kappa index of 0.51 ± 0.17. The results suggest that a good sleep evaluation can be achieved through non-conventional recording systems that could be used outside sleep centers.

Evaluation of the sleep quality based on bed sensor signals: Time-variant analysis.

Automatic detection of the sleep macrostructure (Wake, NREM -non Rapid Eye Movement- and REM -Rapid Eye Movement-) based on bed sensor signals is presented. This study assesses the feasibility of different methodologies to evaluate the sleep quality out of sleep centers. The study compares a) the features extracted from time-variant autoregressive modeling (TVAM) and Wavelet Decomposition (WD) and b) the performance of K-Nearest Neighbor (KNN) and Feed Forward Neural Networks (FFNN) classifiers. In the current analysis, 17 full polysomnography recordings from healthy subjects were used. The best agreement for Wake-NREM-REM with respect to the gold standard was 71.95 ± 7.47% of accuracy and 0.42 ± 0.10 of kappa index for TVAM-LD while WD-FFNN shows 67.17 ± 11.88% of accuracy and 0.39 ± 0.13 of kappa index. The results suggest that the sleep quality assessment out of sleep centers could be possible and as consequence more people could be beneficiated.