Single-slice Alzheimer's disease classification and disease regional analysis with Supervised Switching Autoencoders.

BACKGROUND: Alzheimer's disease (AD) is a difficult to diagnose pathology of the brain that progressively impairs cognitive functions. Computer-assisted diagnosis of AD based on image analysis is an emerging tool to support AD diagnosis. In this article, we explore the application of Supervised Switching Autoencoders (SSAs) to perform AD classification using only one structural Magnetic Resonance Imaging (sMRI) slice. SSAs are revised supervised autoencoder architectures, combining unsupervised representation and supervised classification as one unified model. In this work, we study the capabilities of SSAs to capture complex visual neurodegeneration patterns, and fuse disease semantics simultaneously. We also examine how regions associated to disease state can be discovered by SSAs following a local patch-based approach. METHODS: Patch-based SSAs models are trained on individual patches extracted from a single 2D slice, independently for Axial, Coronal, and Sagittal anatomical planes of the brain at selected informative locations, exploring different patch sizes and network parameterizations. Then, models perform binary class prediction - healthy (CDR = 0) or AD-demented (CDR > 0) - on test data at patch level. The final subject classification is performed employing a majority rule from the ensemble of patch predictions. In addition, relevant regions are identified, by computing accuracy densities from patch-level predictions, and analyzed, supported by Atlas-based regional definitions. RESULTS: Our experiments employing a single 2D T1-w sMRI slice per subject show that SSAs perform similarly to previous proposals that rely on full volumetric information and feature-engineered representations. SSAs classification accuracy on slices extracted along the Axial, Coronal, and Sagittal anatomical planes from a balanced cohort of 40 independent test subjects was 87.5%, 90.0%, and 90.0%, respectively. A top sensitivity of 95.0% on both Coronal and Sagittal planes was also obtained. CONCLUSIONS: SSAs provided well-ranked accuracy performance among previous classification proposals, including feature-engineered and feature learning based methods, using only one scan slice per subject, instead of the whole 3D volume, as it is conventionally done. In addition, regions identified as relevant by SSAs' were, in most part, coherent or partially coherent in regard to relevant regions reported on previous works. These regions were also associated with findings from medical knowledge, which gives value to our methodology as a potential analytical aid for disease understanding.

Voxel-wise assessment of lung aeration changes on CT images using image registration: application to acute respiratory distress syndrome (ARDS).

PURPOSE: (1) To improve the accuracy of global and regional alveolar-recruitment quantification in CT scan pairs by accounting for lung-tissue displacements and deformation, (2) To propose a method for local-recruitment calculation. METHODS: Recruitment was calculated by subtracting the quantity of non-aerated lung tissues between expiration and inspiration. To assess global recruitment, lung boundaries were first interactively delineated at inspiration, and then they were warped based on automatic image registration to define the boundaries at expiration. To calculate regional recruitment, the lung mask defined at inspiration was cut into pieces, and these were also warped to encompass the same tissues at expiration. Local-recruitment map was calculated as follows: For each voxel at expiration, the matching location at inspiration was determined by image registration, non-aerated voxels were counted in the neighborhood of the respective locations, and the voxel count difference was normalized by the neighborhood size. The methods were evaluated on 120 image pairs of 12 pigs with experimental acute respiratory distress syndrome. RESULTS: The dispersion of global- and regional-recruitment values decreased when using image registration, compared to the conventional approach neglecting tissue motion. Local-recruitment maps overlaid onto the original images were visually consistent, and the sum of these values over the whole lungs was very close to the global-recruitment estimate, except four outliers. CONCLUSIONS: Image registration can compensate lung-tissue displacements and deformation, thus improving the quantification of alveolar recruitment. Local-recruitment calculation can also benefit from image registration, and its values can be overlaid onto the original image to display a local-recruitment map. They also can be integrated over arbitrarily shaped regions to assess regional or global recruitment.

Noninvasive quantification of macrophagic lung recruitment during experimental ventilation-induced lung injury.

Macrophagic lung infiltration is pivotal in the development of lung biotrauma because of ventilation-induced lung injury (VILI). We assessed the performance of [11C](R)-PK11195, a positron emission tomography (PET) radiotracer binding the translocator protein, to quantify macrophage lung recruitment during experimental VILI. Pigs (n = 6) were mechanically ventilated under general anesthesia, using protective ventilation settings (baseline). Experimental VILI was performed by titrating tidal volume to reach a transpulmonary end-inspiratory pressure (∆PL) of 35-40 cmH2O. We acquired PET/computed tomography (CT) lung images at baseline and after 4 h of VILI. Lung macrophages were quantified in vivo by the standardized uptake value (SUV) of [11C](R)-PK11195 measured in PET on the whole lung and in six lung regions and ex vivo on lung pathology at the end of experiment. Lung mechanics were extracted from CT images to assess their association with the PET signal. ∆PL increased from 9 ± 1 cmH2O under protective ventilation, to 36 ± 6 cmH2O during experimental VILI. Compared with baseline, whole-lung [11C](R)-PK11195 SUV significantly increased from 1.8 ± 0.5 to 2.9 ± 0.5 after experimental VILI. Regional [11C](R)-PK11195 SUV was positively associated with the magnitude of macrophage recruitment in pathology (P = 0.03). Compared with baseline, whole-lung CT-derived dynamic strain and tidal hyperinflation increased significantly after experimental VILI, from 0.6 ± 0 to 2.0 ± 0.4, and 1 ± 1 to 43 ± 19%, respectively. On multivariate analysis, both were significantly associated with regional [11C](R)-PK11195 SUV. [11C](R)-PK11195 lung uptake (a proxy of lung inflammation) was increased by experimental VILI and was associated with the magnitude of dynamic strain and tidal hyperinflation.NEW & NOTEWORTHY We assessed the performance of [11C](R)-PK11195, a translocator protein-specific positron emission tomography (PET) radiotracer, to quantify macrophage lung recruitment during experimental ventilation-induced lung injury (VILI). In this proof-of-concept study, we showed that the in vivo quantification of [11C](R)-PK11195 lung uptake in PET reflected the magnitude of macrophage lung recruitment after VILI. Furthermore, increased [11C](R)-PK11195 lung uptake was associated with harmful levels of dynamic strain and tidal hyperinflation applied to the lungs.

A tree-matching algorithm: Application to airways in CT images of subjects with the acute respiratory distress syndrome.

To match anatomical trees such as airways, we propose a graph-based strategy combined with an appropriate distance function. The strategy was devised to cope with topological and geometrical differences that may arise between trees corresponding to the same subject, but extracted from images acquired in different conditions. The proposed distance function, called father/family distance, combines topological and geometrical information in a single measure, by calculating a sum of path-to-path distances between sub-trees of limited extent. To use it successfully, the branches of these sub-trees need to be brought closer, which is obtained by successively translating the roots of these sub-trees prior to their actual matching. The work herein presented contributes to a study of the acute respiratory distress syndrome, where a series of pulmonary CT images from the same subject is acquired at varying settings (pressure and volume) of the mechanical ventilation. The method was evaluated on 45 combinations of synthetic trees, as well as on 15 pairs of real airway trees: nine corresponding to end-expiration and end-inspiration with the same pressure, and six corresponding to end-inspiration with significantly different pressures. It achieved a high rate of successful matches with respect to a hand-made reference containing a total of 2391 matches in real data: sensitivity of 94.3% and precision of 92.8%, when using the basic parameter settings of the algorithm.

A semi-automatic method to extract canal pathways in 3D micro-CT images of Octocorals.

The long-term goal of our study is to understand the internal organization of the octocoral stem canals, as well as their physiological and functional role in the growth of the colonies, and finally to assess the influence of climatic changes on this species. Here we focus on imaging tools, namely acquisition and processing of three-dimensional high-resolution images, with emphasis on automated extraction of canal pathways. Our aim was to evaluate the feasibility of the whole process, to point out and solve - if possible - technical problems related to the specimen conditioning, to determine the best acquisition parameters and to develop necessary image-processing algorithms. The pathways extracted are expected to facilitate the structural analysis of the colonies, namely to help observing the distribution, formation and number of canals along the colony. Five volumetric images of Muricea muricata specimens were successfully acquired by X-ray computed tomography with spatial resolution ranging from 4.5 to 25 micrometers. The success mainly depended on specimen immobilization. More than [Formula: see text] of the canals were successfully detected and tracked by the image-processing method developed. Thus obtained three-dimensional representation of the canal network was generated for the first time without the need of histological or other destructive methods. Several canal patterns were observed. Although most of them were simple, i.e. only followed the main branch or "turned" into a secondary branch, many others bifurcated or fused. A majority of bifurcations were observed at branching points. However, some canals appeared and/or ended anywhere along a branch. At the tip of a branch, all canals fused into a unique chamber. Three-dimensional high-resolution tomographic imaging gives a non-destructive insight to the coral ultrastructure and helps understanding the organization of the canal network. Advanced image-processing techniques greatly reduce human observer's effort and provide methods to both visualize and quantify the structures of interest.

Creatools: una plataforma para el desarrollo de software de procesamiento de imágenes médicas. Aplicación a la segmentación de arterias coronarias y a la detección automática de lesiones en imágenes de angiotac / Creatools: a framework for the development of medical image processing software. Application for the segmentation of coronary arteries and for the automatic detection of lesions in CT images

En este artículo se presenta un software de código abierto, llamado CreaTools, cuyo principal objetivo es el procesar y facilitar la visualización de imágenes médicas. Este software flexible funciona en diferentes sistemas operativos (Linux, Mac OS X, Windows), se desarrolla en el lenguaje de programación C++ para asegurar una fácil integración de módulos C++ y proporciona a los usuarios herramientas computacionales para construir interfaces gráficas de usuario (GUI), incluidos los datos de entrada/salida (manejo de archivos), la visualización, la interacción y el procesamiento de datos. Este artículo muestra también la utilidad de CreaTools mediante un proyecto de investigación que consiste en la detección automática de lesiones arteriales. Los algoritmos desarrollados han sido implementados en una interfaz gráfica amigable con visualización 3D e interacción. Ejemplos de tales algoritmos incluyen la extracción de ejes de arterias y la generación de modelos descriptivos de arterias con lesiones y sin lesiones.

Learning from only positive and unlabeled data to detect lesions in vascular CT images.

Detecting vascular lesions is an important task in the diagnosis and follow-up of the coronary heart disease. While most existing solutions tackle calcified and non-calcified plaques separately, we present a new algorithm capable of detecting both types of lesions in CT images. It builds up on a semi-supervised classification framework, in which the training set is made of both unlabeled data and a small amount of data labeled as normal. Our method takes advantage of the arrival of newly acquired data to re-train the classifier and improve its performance. We present results on synthetic data and on datasets from 15 patients. With a small amount of labeled training data our method achieved a 89.8% true positive rate, which is comparable to state-of-the-art supervised methods, and the performance can improve after additional iterations.

Automatic detection of abnormal vascular cross-sections based on density level detection and support vector machines.

PURPOSE: The goal is to automatically detect anomalous vascular cross-sections to attract the radiologist's attention to possible lesions and thus reduce the time spent to analyze the image volume. MATERIALS AND METHODS: We assume that both lesions and calcifications can be considered as local outliers compared to a normal cross-section. Our approach uses an intensity metric within a machine learning scheme to differentiate normal and abnormal cross-sections. It is formulated as a Density Level Detection problem and solved using a Support Vector Machine (DLD-SVM). The method has been evaluated on 42 synthetic phantoms and on 9 coronary CT data sets annotated by 2 experts. RESULTS: The specificity of the method was 97.57% on synthetic data, and 86.01% on real data, while its sensitivity was 82.19 and 81.23%, respectively. The agreement with the observers, measured by the kappa coefficient, was substantial (κ = 0.72). After the learning stage, which is performed off-line, the average processing time was within 10 s per artery. CONCLUSIONS: To our knowledge, this is the first attempt to use the DLD-SVM approach to detect vascular abnormalities. Good specificity, sensitivity and agreement with experts, as well as a short processing time, show that our method can facilitate medical diagnosis and reduce evaluation time by attracting the reader's attention to suspect regions.

Adaptación del llgoritmo maracas para segmentacón de la arteria carótida y cuaantificación de estenosis en imágenes aac / Adaptation of the MARACAS Algorithm for Carotid Artery Segmentation and Stenosis Quantification on CT Images

En este artículo se describen las adaptaciones hechas al algoritmo MARACAS para segmentar y cuantificar estructuras vasculares en imágenes TAC de la arteria carótida. El algoritmo MARACAS, que está basado en un modelo elástico y en un análisis de los valores y vectores propios de la matriz de inercia, fue inicialmente diseñado para segmentar una sola arteria en imágenes ARM. Las modificaciones están principalmente enfocadas a tratar las especificidades de las imágenes TAC, así como la presencia de bifurcaciones. Los algoritmos implementados en esta nueva versión se clasifican en dos niveles. 1. Los procesamientos de bajo nivel (filtrado de ruido y de artificios direccionales, presegmentación y realce) destinados a mejorar la calidad de la imagen y presegmentarla. Estas técnicas están basadas en información a priori sobre el ruido, los artificios y los intervalos típicos de niveles de gris del lumen, del fondo y de las calcificaciones. 2. Los procesamientos de alto nivel para extraer la línea central de la arteria, segmentar el lumen y cuantificar la estenosis. A este nivel, se aplican conocimientos a priori sobre la forma y anatomía de las estructuras vasculares. El método fue evaluado en 31 imágenes suministradas en el concurso Carotid Lumen Segmentation and Stenosis Grading Grand Challenge 2009. Los resultados obtenidos en la segmentación arrojaron un coeficiente de similitud de Dice promedio de 80,4% comparado con la segmentación de referencia, y el error promedio de la cuantificación de estenosis fue 14,4%.

This paper describes the adaptations of MARACAS algorithm to the segmentation and quantification of vascular structures in CTA images of the carotid artery. The MARACAS algorithm, which is based on an elastic model and on a multi-scale eigen-analysis of the inertia matrix, was originally designed to segment a single artery in MRA images. The modifications are primarily aimed at addressing the specificities of CT images and the bifurcations. The algorithms implemented in this new version are classified into two levels. 1. The low-level processing (filtering of noise and directional artifacts, enhancement and pre-segmentation) to improve the quality of the image and to pre-segment it. These techniques are based on a priori information about noise, artifacts and typical gray levels ranges of lumen, background and calcifications. 2. The high-level processing to extract the centerline of the artery, to segment the lumen and to quantify the stenosis. At this level, we apply a priori knowledge of shape and anatomy of vascular structures. The method was evaluated on 31 datasets from the Carotid Lumen Segmentation and Stenosis Grading Grand Challenge 2009. The segmentation results obtained an average of 80:4% Dice similarity score, compared to reference segmentations, and the mean stenosis quantification error was 14.4%.

Mejoramiento de la segmentación arterial coronaria a través del modelamiento del volumen parcial en imágenes de tomografía computarizada / Coronar y arter y segmentation improvement thro ugh partial volume modeling in CT images

En este artículo se propone un modelo estadístico de volumen parcial (VP) para mejorar la segmentación 3D de imágenes de tomografía computarizada (TC) cardiaca. Los efectos causados por el VP representan un reto en la separación arterial de las cavidades cardiacas, porque causan desbordamientos y segmentaciones erróneas. La propuesta incluye un campo aleatorio de Markov junto con un esquema de pesos modificado. Además, se utilizaron fantasmas sintéticos para evaluar la precisión del método, así como para determinar los parámetros de configuración ideales. Se usaron las imágenes de ocho pacientes, a fin de evaluar el método sobre datos reales, y se comparó el desempeño del esquema de pesos modificado con el esquema tradicional. También se demostró la capacidad del método para mejorar la segmentación cuando se usa en conjunto con un algoritmo de extracción de la línea central arterial.

In this article it is proposed a statistic model of Partial Volume (VP in spanish) to improve the 3D segmentation of Heart CT images. The effects caused by the VP represent a challenge in the arterial separation of the heart cavities because they cause overflowing and wrong segmentations. The proposal includes a random Markov field along to a modified weight scheme. Besides, synthetic ghosts where used to asses the precision of the method as well as to determine the parameters of ideal settings. The images of eight patients were used to evaluate the method based on real data and the performance of the modified weight scheme was compared with the traditional scheme. The ability of the method to improve the segmentation was proved when it was used along with a central arterial line extraction algorithm.

Standardized evaluation methodology and reference database for evaluating coronary artery centerline extraction algorithms.

Efficiently obtaining a reliable coronary artery centerline from computed tomography angiography data is relevant in clinical practice. Whereas numerous methods have been presented for this purpose, up to now no standardized evaluation methodology has been published to reliably evaluate and compare the performance of the existing or newly developed coronary artery centerline extraction algorithms. This paper describes a standardized evaluation methodology and reference database for the quantitative evaluation of coronary artery centerline extraction algorithms. The contribution of this work is fourfold: (1) a method is described to create a consensus centerline with multiple observers, (2) well-defined measures are presented for the evaluation of coronary artery centerline extraction algorithms, (3) a database containing 32 cardiac CTA datasets with corresponding reference standard is described and made available, and (4) 13 coronary artery centerline extraction algorithms, implemented by different research groups, are quantitatively evaluated and compared. The presented evaluation framework is made available to the medical imaging community for benchmarking existing or newly developed coronary centerline extraction algorithms.

Modelamiento en 3D de patologías vasculares a partir de imágenes de resonancia magnética sustraída / 3D Modeling of Vascular Pathologies from contrast enhanced magnetic resonance images (MRI)

Este artículo presenta un método para la generación de modelos vasculares en 3D, a partir de imágenes de resonancia magnética (IRM), usando un algoritmo de fast marching. Los principales aportes del método propuesto en este artículo son la utilización de la imagen original como base para la definición de la función de velocidad que rige el desplazamiento de la interfaz y la selección automática del tiempo en el cual la interfaz logra segmentar la arteria. El método fue validado en imágenes de arterias carótidas patológicas y de fantasmas vasculares. Una apreciación cualitativa de los modelos vasculares obtenidos muestra una extracción adecuada de la pared vascular. Una validación cuantitativa demostró que los modelos generados dependen de la escogencia de los parámetros del algoritmo, al inducir un error máximo de 1,34 vóxeles en el diámetro de las estenosis medidas.

Segmentación y visualización arterial en ARM y TAC / Segmentation and arterial visualization in MRA and CT

Este artículo describe un método para la extracción automática de estructuras vasculares en imágenes médicas en 3D. El método utiliza un algoritmo iterativo que adiciona puntos al esqueleto del vaso y detecta bifurcaciones que analizan el contenido de una esfera que se mueve a lo largo de su línea central. En cada iteración se realiza una segmentación (extracción del vaso) local dentro de la esfera. Esta acción emplea un algoritmo de K-medias, que separa vaso y fondo utilizando métricas diferentes para cada grupo; adicionalmente, una medida del cilindricidad, basada en la comparación del volumen segmentado contra un modelo construido del vaso, se usa como el criterio de parada del algoritmo. El método fue aplicado a 16 ARM y a 12 TC 3D de diversas regiones anatómicas: arterias carótidas, árbol pulmonar, arterias coronarias y aorta. El algoritmo detectó y manejó eficientemente las bifurcaciones. Cada imagen fue procesada en menos de cinco minutos, lo cual es bastante rápido como para ser utilizada en una rutina clínica.

Segmentación y caracterización de la arteria carótida en escanografías 3d / Segmentation and characterization of the carothid artery in 3d scanners

Este artículo presenta un método de segmentación vascular y caracterización de placas ateroscleróticas en imágenes de tomografía computarizada 3D. El primer paso hacia este objetivo es la extracción de la línea central de la arteria por medio de un método de esqueleto extensible. Este método utiliza un esquema estimaciónpredicción iterativo, análisis multiescala de momentos de la imagen y un modelo de forma de segundo orden. Los contornos vasculares y de placas son detectados en una segunda etapa sobre los planos localmente perpendiculares a la línea central. Los puntos de los contornos están determinados por una búsqueda de los máximos locales del gradiente de intensidad, calculado en direcciones radiales a partir del punto del eje central de la arteria. Resultados experimentales son presentados sobre imágenes diagnósticas 3D de arterias carótidas patológicas.

Standardized MR protocol for the evaluation of MRA sequences and/or contrast agents effects in high-degree arterial stenosis analysis.

PURPOSE: To investigate the relative role of high resolution (spatial or temporal) magnetic resonance angiography (MRA) sequence and of contrast agent properties in the evaluation of high-degree arterial stenosis. METHODS: We qualitatively and quantitatively studied both 50 and 95% (300 microm diameter) stenosis of a 6 mm arterial phantom with two contrast agents (CA), Gd-DOTA (r(1)=2.9 mM(-1) s(-1)) versus P760 (r(1)=25 mM(-1) s(-1)) at several CA concentrations, including arterial peak concentration after injection of either a single or double dose of CA, using either a high temporal (booster) or high spatial (HR) resolution 3D MRA sequences. Experimental data were then compared to theoretical data. RESULTS: With the 3D HR sequence, both visual and quantitative analysis were significantly better compared to the 3D booster sequence, at each phantom diameter. Quantitative analysis was significantly improved by injection of a double versus a single dose of each CA (Gd-DOTA or P760), primarily in high degree stenosis. CONCLUSION: Combined MRA spatial resolution and high CA efficiency are mandatory to correctly evaluate high degree stenosis.

Computer-assisted analysis of three-dimensional MR angiograms.

The software tools required for postprocessing of magnetic resonance (MR) angiograms include the following functions: data handling, image visualization, and vascular analysis. A custom postprocessing software called Magnetic Resonance Angiography Computer Assisted Analysis (MARACAS) has been developed. This software combines the most commonly used three-dimensional visualization techniques with image processing methods for analysis of vascular morphology on MR angiograms. The main contributions of MARACAS are (a) implementation of a fast method for stenosis quantification on three-dimensional MR angiograms, which is clinically applicable in a personal computer-based system; and (b) portability to the most widespread platforms. The quantification is performed in three steps: extraction of the vessel centerline, detection of vessel boundaries in planes locally orthogonal to the centerline, and calculation of stenosis parameters on the basis of the resulting contours. Qualitative results from application of the method to data from patients showed that the vessel centerline correctly tracked the vessel path and that contours were correctly estimated. Quantitative results obtained from images of phantoms showed that the computation of stenosis severity was accurate.