Search

Abstract The use of prior knowledge in the machine learning framework has been considered a potential tool to handle the curse of dimensionality in genetic and genomics data. Although random forest (RF) represents a flexible non-parametric approach with several advantages, it can provide poor accuracy in high-dimensional settings, mainly in scenarios with small sample sizes. We propose a knowledge-slanted RF that integrates biological networks as prior knowledge into the model to improve its performance and explainability, exemplifying its use for selecting and identifying relevant genes. knowledge-slanted RF is a combination of two stages. First, prior knowledge represented by graphs is translated by running a random walk with restart algorithm to determine the relevance of each gene based on its connection and localization on a protein-protein interaction network. Then, each relevance is used to modify the selection probability to draw a gene as a candidate split-feature in the conventional RF. Experiments in simulated datasets with very small sample sizes $$(n \le 30)$$ ( n ≤ 30 ) comparing knowledge-slanted RF against conventional RF and logistic lasso regression, suggest an improved precision in outcome prediction compared to the other methods. The knowledge-slanted RF was completed with the introduction of a modified version of the Boruta feature selection algorithm. Finally, knowledge-slanted RF identified more relevant biological genes, offering a higher level of explainability for users than conventional RF. These findings were corroborated in one real case to identify relevant genes to calcific aortic valve stenosis.

Las dosis de radiación ionizante en tomografía computarizada (TC) tienen el potencial de generar efectos deletéreos. El tema no es trivial y la pregunta permanece en cuanto a cómo se sitúa el manejo de dosis, en concreto, en Chile. El objetivo de este trabajo es registrar y sintetizar las dosis por paciente en examen de TC en tres instituciones de salud de la Región de Valparaíso. Se realizó un estudio prospectivo descriptivo durante 3 meses, incluyendo exámenes de adultos de cabeza, tórax, abdomen y pelvis. Se recolectaron archivos DICOM asociados para extraer valores de variables relevantes. Se observaron distribuciones de dosis diferentes entre hospitales, al tener equipos y protocolos diferentes. Las medianas de CTDI (Computed Tomography Dose Index) observadas están en general dentro del rango internacional, pero con una gran proporción de exámenes que sobrepasan estos valores. Las variables de mayor relevancia en el control de la dosis fueron el número de series asociado a cada examen y la corriente de tubo. Este estudio muestra que las dosis a pacientes en estas tres instituciones son conformes a lo reportado en el ámbito internacional, con cierto espacio de mejora existente. Queda por confirmarse si estas conclusiones se pueden extender a otras instituciones y otras regiones de Chile.

Medical image quality is crucial to obtaining reliable diagnostics. Most quality controls rely on routine tests using phantoms, which do not reflect closely the reality of images obtained on patients and do not reflect directly the quality perceived by radiologists. The purpose of this work is to develop a method that classifies the image quality perceived by radiologists in MR images. The focus was set on lumbar images as they are widely used with different challenges. Three neuroradiologists evaluated the image quality of a dataset that included T1-weighting images in axial and sagittal orientation, and sagittal T2-weighting. In parallel, we introduced the computational assessment using a wide range of features extracted from the images, then fed them into a classifier system. A total of 95 exams were used, from our local hospital and a public database, and part of the images was manipulated to broaden the distribution quality of the dataset. Good recall of 82% and an area under curve (AUC) of 77% were obtained on average in testing condition, using a Support Vector Machine. Even though the actual implementation still relies on user interaction to extract features, the results are promising with respect to a potential implementation for monitoring image quality online with the acquisition process.

Cerebral aneurysm is a cerebrovascular disorder characterized by a bulging in a weak area in the wall of an artery that supplies blood to the brain. It is relevant to understand the mechanisms leading to the apparition of aneurysms, their growth and, more important, leading to their rupture. The purpose of this study is to study the impact on aneurysm rupture of the combination of different parameters, instead of focusing on only one factor at a time as is frequently found in the literature, using machine learning and feature extraction techniques. This discussion takes relevance in the context of the complex decision that the physicians have to take to decide which therapy to apply, as each intervention bares its own risks, and implies to use a complex ensemble of resources (human resources, OR, etc.) in hospitals always under very high work load. This project has been raised in our actual working team, composed of interventional neuroradiologist, radiologic technologist, informatics engineers and biomedical engineers, from Valparaiso public Hospital, Hospital Carlos van Buren, and from Universidad de Valparaíso – Facultad de Ingeniería and Facultad de Medicina. This team has been working together in the last few years, and is now participating in the implementation of an “interdisciplinary platform for innovation in health”, as part of a bigger project leaded by Universidad de Valparaiso (PMI UVA1402). It is relevant to emphasize that this project is made feasible by the existence of this network between physicians and engineers, and by the existence of data already registered in an orderly manner, structured and recorded in digital format. The present proposal arises from the description in nowadays literature that the actual indicators, whether based on morphological description of the aneurysm, or based on characterization of biomechanical factor or others, these indicators were shown not to provide sufficient information in order to predict by themselves the risk of rupture. Therefore, our hypothesis is that the risk of rupture lies on the combination of multiple actors. These actors together would play different roles that could be: weakening of the artery wall, increasing biomechanical stresses on the wall induced by blood flow, in addition to personal sensitivity due to family history, or personal history of comorbidity, or even seasonal variations that could gate different inflammation mechanisms. The main goal of this project is to identify relevant variables that may help in the process of predicting the risk of intracranial aneurysm rupture using machine learning and image processing techniques based on structured and non-structured data from multiple sources. We believe that the identification and the combined use of relevant variables extracted from clinical, demographical, environmental and medical imaging data sources will improve the estimation of the aneurysm rupture risk, with respect to the actual practiced method based essentially on the aneurysm size. The methodology of this work consist of four phases: (1) Data collection and storage, (2) feature extraction from multiple sources in particular from angiographic images, (3) development of the model that could describe the risk of aneurysm rupture based on the fusion and combination of the features, and (4) Identification of relevant variables related to the aneurysm rupture process. This study corresponds to an analytic transversal study with prospective and retrospective characteristics. This work will be based on publicly available health statistics data, data of weather conditions, together with clinical and demographic data of patients diagnosed with intracranial aneurysm in the Hospital Carlos van Buren. As main results of this project we are expecting to identify relevant variables extracted from images and other sources that could play a role in the risk of aneurysm rupture. The proposed model will be presented to the physicians of the Hospital Carlos van Buren, to be further implemented in this Institution according to the demonstrated impact of our results. The main results will be published in indexed journals and presented at national and international conferences.

Objetivo: Los parámetros hemodinámicos son de gran utilidad para realizar un adecuado diagnóstico. Sin embargo, debido a la gran cantidad de variables que pueden obtenerse, el análisis global de todas ellas puede ser complejo. Para facilitar esta tarea, nosotros proponemos crear un modelo que permita clasificar distintas variables hemodinámicas entre las pertenecientes a un individuo sano o a uno patológico. Para ello, usaremos técnicas de minería de datos que permitan identificar y encontrar relaciones entre distintos parámetros hemodinámicos de la aorta obtenidos a través de flujo multidimensional (4D flow) por resonancia magnética. Método: Una secuencia 4D flow de todo el corazón y los grandes vasos fue adquirida utilizando resonancia magnética en 19 voluntarios sanos y 2 pacientes (uno con una coartación aórtica y otro con una coartación aórtica reparada). Retrospectivamente, los datos fueron reformateados a lo largo de la aorta, originándose 3 cortes en los voluntarios y 30 cortes en cada paciente. En cada corte la aorta fue segmentada y distintos parámetros fueron cuantificados: área, velocidad máxima, velocidad mínima, flujo y volumen, calculándose en los cuatro últimos su valor máximo, promedio, desviación estándar, curtosis, sesgo, proporción de tiempo en alcanzar el valor máximo, entre otros. Teniendo un total de 26 variables por cada corte. Se aplicó la técnica de árboles de decisión tipo CART (por sus siglas en inglés) para clasificar los datos. Para validar el modelo, 2 cortes extras fueron generados por cada voluntario y 20 cortes por cada paciente. Resultados: La técnica CART, mediante la utilización de sólo 7 variables, puede clasificar las imágenes de los voluntarios y pacientes con una tasa de error del 14,1%, una sensibilidad de 82,5% y una especificidad de 89.4%. Conclusiones: 4D flow provee una gran cantidad de datos hemodinámicos que son difíciles de analizar. En este trabajo demostramos que al utilizar minería de datos se pueden clasificar imágenes a partir de parámetros hemodinámicos relevantes y sus relaciones para apoyar el diagnóstico de alteraciones cardiovasculares.Objective: Hemodynamic parameters are critical to perform a proper diagnosis. However, due to the large number of variables that can be obtained, overall analysis may represent a complex task. To facilitate this, we propose to create a model for classifying different hemodynamic variables between those belonging to a healthy individual and to a pathological patient. For this purpose, we employed data mining techniques to identify relationships among various aortic hemodynamic parameters obtained through multi-dimensional (4D flow) MR imaging. Method: A 4D flow sequence of whole heart and great vessels was acquired using MRI in 19 healthy volunteers and 2 patients (one with aortic coarctation and one with repaired coarctation of the aorta). Retrospectively, data were reformatted along the aorta; three MRI acquisitions were performed for volunteers and 30 sequences for each patient. In each slice the aorta was segmented and various parameters were quantified: area, maximum velocity, minimum velocity, flow and volumen, with following values being calculated for last four parameters: maximum, average, standard deviation, kurtosis, skewness, proportion of time to reach the maximum value, among others. A total of 26 variables for each acquisition were obtained. In order to classify data, the CART Technique (Classification and Regression Trees) was applied. To validate the model, two extra projections were generated per each volunteer and 20 slice per each patient. Results: By using only 7 variables, the CART Technique allows discrimination between images performed either on volunteers or patients with an error rate of 14.1%, a sensitivity of 82.5%, and a specificity of 89.4%. Conclusions: 4D flow MR imaging provides a wealth of hemodynamic data that can be difficult to analyze. In this paper we demonstrate that by using data mining techniques it is possible to classify images from relevant hemodynamic parameters and their relationships in order to support the diagnosis of cardiovascular disorders.

Linear functional analysis historically founded by Fourier and Legendre played a significant role to provide a unified vision of mathematical transformations between vector spaces. The possibility of extending this approach is explored when basis of vector spaces is built Tailored to the Problem Specificity (TPS) and not from the convenience or effectiveness of mathematical calculations. Standardized mathematical transformations, such as Fourier or polynomial transforms, could be extended toward TPS methods, on a basis, which properly encodes specific knowledge about a problem. Transition between methods is illustrated by comparing what happens in conventional Fourier transform with what happened during the development of Jewett Transform, reported in previous articles. The proper use of computational intelligence tools to perform Jewett Transform allowed complexity algorithm optimization, which encourages the search for a general TPS methodology.

Menopause transition can be interpreted as a vulnerable state characterized by estrogen deficiency with detrimental systemic effects as the low-grade chronic inflammation that appears with aging and partly explains age-related disorders as cancer, diabetes mellitus and increased risk of cognitive impairment. Over the course of a lifetime, estrogen produces several beneficial effects in healthy neurological tissues as well as cardioprotective effects, and anti-inflammatory effects. However, clinical evidence on the efficacy of hormone treatment in menopausal women has failed to confirm the benefit reported in observational studies. Unambiguously, enhanced verbal memory is the most robust finding from longitudinal and cross-sectional studies, what merits consideration for future studies aiming to determine estrogen neuroprotective efficacy. Estrogen related brain activity and functional connectivity remain, however, unexplored. In this context, the resting state paradigm may provide valuable information about reproductive aging and hormonal treatment effects, and their relationship with brain imaging of functional connectivity may be key to understand and anticipate estrogen cognitive protective effects. To go in-depth into the molecular and cellular mechanisms underlying rapid-to-long lasting protective effects of estrogen, we will provide a comprehensive review of cognitive tasks used in animal studies to evaluate the effect of hormone treatment on cognitive performance and discuss about the tasks best suited to the demonstration of clinically significant differences in cognitive performance to be applied in human studies. Eventually, we will focus on studies evaluating the DMN activity and responsiveness to pharmacological stimulation in humans.

Introduction: In Chile, 1 in 8 pregnant women of middle socioeconomic level has gestational diabetes mellitus (GDM), and in general, 5–10% of women with GDM develop type 2 diabetes after giving birth. Recently, various technological tools have emerged to assist patients with GDM to meet glycemic goals and facilitate constant glucose monitoring, making these tasks more straightforward and comfortable.Objective: To evaluate the impact of remote monitoring technologies in assisting patients with GDM to achieve glycemic goals, and know the respective advantages and disadvantages when it comes to reducing risk during pregnancy, both for the mother and her child.Methods: A total of 188 articles were obtained with the keywords “gestational diabetes mellitus,” “GDM,” “gestational diabetes,” added to the evaluation levels associated with “glucose level,” “glycemia,” “glycemic index,” “blood sugar,” and the technological proposal to evaluate with “glucometerm” “mobile application,” “mobile applications,” “technological tools,” “telemedicine,” “technovigilance,” “wearable” published during the period 2016–2021, excluding postpartum studies, from three scientific databases: PUBMED, Scopus and Web of Science. These were managed in the Mendeley platform and classified using the PRISMA method.Results: A total of 28 articles were selected after elimination according to inclusion and exclusion criteria. The main measurement was glycemia and 4 medical devices were found (glucometer: conventional, with an infrared port, with Bluetooth, Smart type and continuous glucose monitor), which together with digital technology allow specific functions through 2 identified digital platforms (mobile applications and online systems). In four articles, the postprandial glucose was lower in the Tele-GDM groups than in the control group. Benefits such as improved glycemic control, increased satisfaction and acceptability, maternal confidence, decreased gestational weight gain, knowledge of GDM, and other relevant aspects were observed. There were also positive comments regarding the optimization of the medical team’s time.Conclusion: The present review offers the opportunity to know about the respective advantages and disadvantages of remote monitoring technologies when it comes to reducing risk during pregnancy. GDM centered technology may help to evaluate outcomes and tailor personalized solutions to contribute to women’s health. More studies are needed to know the impact on a healthcare system.

Introduction: Artificial intelligence is widely used in medical field, and machine learning has been increasingly used in health care, prediction, and diagnosis and as a method of determining priority. Machine learning methods have been features of several tools in the fields of obstetrics and childcare. This present review aims to summarize the machine learning techniques to predict perinatal complications.Objective: To identify the applicability and performance of machine learning methods used to identify pregnancy complications.Methods: A total of 98 articles were obtained with the keywords “machine learning,” “deep learning,” “artificial intelligence,” and accordingly as they related to perinatal complications (“complications in pregnancy,” “pregnancy complications”) from three scientific databases: PubMed, Scopus, and Web of Science. These were managed on the Mendeley platform and classified using the PRISMA method.Results: A total of 31 articles were selected after elimination according to inclusion and exclusion criteria. The features used to predict perinatal complications were primarily electronic medical records (48%), medical images (29%), and biological markers (19%), while 4% were based on other types of features, such as sensors and fetal heart rate. The main perinatal complications considered in the application of machine learning thus far are pre-eclampsia and prematurity. In the 31 studies, a total of sixteen complications were predicted. The main precision metric used is the AUC. The machine learning methods with the best results were the prediction of prematurity from medical images using the support vector machine technique, with an accuracy of 95.7%, and the prediction of neonatal mortality with the XGBoost technique, with 99.7% accuracy.Conclusion: It is important to continue promoting this area of research and promote solutions with multicenter clinical applicability through machine learning to reduce perinatal complications. This systematic review contributes significantly to the specialized literature on artificial intelligence and women’s health.

Abstract Background Education and health are crucial topics for public policies as both largely determine the future wellbeing of the society. Currently, several studies recognize that physical activity (PA) benefits brain health in children. However, most of these studies have not been carried out in developing countries or lack the transference into the education field. The Cogni-Action Project is divided into two stages, a cross-sectional study and a crossover-randomized trial. The aim of the first part is to establish the associations of PA, sedentarism, and physical fitness with brain structure and function, cognitive performance and academic achievement in Chilean schoolchildren (10–13 years-old). The aim of the second part is to determinate the acute effects of three PA protocols on neuroelectric indices during a working memory and a reading task. Methods PA and sedentarism will be self-reported and objectively-assessed with accelerometers in a representative subsample, whilst physical fitness will be evaluated through the ALPHA fitness test battery. Brain structure and function will be assessed by magnetic resonance imaging (MRI) in a randomized subsample. Cognitive performance will be assessed through the NeuroCognitive Performance Test, and academic achievement by school grades. In the second part 32 adolescents (12–13 year-old) will be cross-over randomized to these condition (i) “Moderate-Intensity Continuous Training” (MICT), (ii) “Cooperative High-Intensity Interval Training” (C-HIIT), and (iii) Sedentary condition. Neuroelectric indices will be measures by electroencephalogram (EEG) and eye-tracking, working memory by n-back task and reading comprehension by a reading task. Discussion The main strength of this project is that, to our knowledge, this is the first study analysing the potential association of PA, sedentarism, and physical fitness on brain structure and function, cognitive performance, and academic achievement in a developing country, which presents an important sociocultural gap. For this purpose, this project will use advanced technologies in neuroimaging (MRI), electrophysiology (EEG), and eye-tracking, as well as objective and quality measurements of several physical and cognitive health outcomes. Trial registration ClinicalTrials.gov identifier: NCT03894241 Date of register: March 28, 2019. Retrospectively Registered.

Deep learning models are part of the family of artificial neural networks and, as such, they suffer catastrophic interference when learning sequentially. In addition, the greater number of these models have a rigid architecture which prevents the incremental learning of new classes. To overcome these drawbacks, we propose the Self-Improving Generative Artificial Neural Network (SIGANN), an end-to-end deep neural network system which can ease the catastrophic forgetting problem when learning new classes. In this method, we introduce a novel detection model that automatically detects samples of new classes, and an adversarial autoencoder is used to produce samples of previous classes. This system consists of three main modules: a classifier module implemented using a Deep Convolutional Neural Network, a generator module based on an adversarial autoencoder, and a novelty-detection module implemented using an OpenMax activation function. Using the EMNIST data set, the model was trained incrementally, starting with a small set of classes. The results of the simulation show that SIGANN can retain previous knowledge while incorporating gradual forgetfulness of each learning sequence at a rate of about 7% per training step. Moreover, SIGANN can detect new classes that are hidden in the data with a median accuracy of 43 % and, therefore, proceed with incremental class learning.

Nuestro objetivo fue implementar un protocolo de estimulación en resonancia magnética funcional en un Hospital de Valparaíso, V Región de Chile, para detectar las áreas corticales activadas en la comprensión del lenguaje oral. Siete voluntarios sanos participaron de este estudio. El t-score y variación de señal alcanzado fue de 6.3 0.3 y 0.5±0.1% respectivamente. A pesar de estos bajos valores, las activaciones se registraron en el área Wernicke, circunvolución temporal media y circunvolución d e Heschl. La extensión de las activaciones fue pequeña, 5.06±2.99 cm³, probablemente debido a la baja relación señal ruido del resonador (SNR), además de la complejidad cognitiva de la tarea y el ruido acústico ambiente. Nuestros resultados indican que la implementación de la RMf en comprensión del lenguaje es posible de realizar en Chile sin recursos adicionales.This study aimed at implementing a stimulation protocol using functional Magnetic Resonance Imaging (fMRI), in a Hospital in Valparaiso (V Region), Chile, to detect cortical areas activated in oral language comprehension. Seven healthy volunteers participated in this study. Average t-score and signal variation were 6.3±0.3 and 0.5±0.1%, respectively. In spite of these low values, activations were obtained in Wernicke area, middle temporal gyrus, and Heschl s gyrus. The extension of activated areas was small, 5.06±2.99 cm3, probably due to amplifer low signal-to-noise ratio (SNR), in addition to the cognitive complexity of the task, and to the ambient acoustic noise. Successful implementation of fMRI protocols of language comprehension is possible in a clinical context in Chile without any additional resources.

Diffusion Magnetic Resonance Imaging provides images of unquestionable diagnostic value. It is commonly used in the assessment of stroke and in white matter fiber tracking, among other applications. The diffusion coefficient has been shown to depend on cell concentration, membrane permeability, and cell orientation in the case of white matter or muscle fiber tracking; yet a clear relation between diffusion measurements and known physiological parameters is not established. The aim of this paper is to review hypotheses and actual knowledge on diffusion signal origin to provide assistance in the interpretation of diffusion MR images. Focus will be set on brain images, as most common applications of diffusion MRI are found in neuroradiology. Diffusion signal does not come from two intra- or extracellular compartments, as was first assumed. Restriction of water displacement due to membranes, hindrance in the extracellular space, and tissue heterogeneity are important factors. Unanswered questions remain on how to deal with tissue heterogeneity, and how to retrieve parameters less troublesome to work with from biological and clinical points of view. Diffusion quantification should be done with care, as many variables can lead to variation in measurements

Nuestro objetivo fue implementar un protocolo de estimulación en resonancia magnética funcional en un Hospital de Valparaíso, V Región de Chile, para detectar las áreas corticales activadas en la comprensión del lenguaje oral. Siete voluntarios sanos participaron de este estudio. El t-score y variación de señal alcanzado fue de 6.3 0.3 y 0.5±0.1% respectivamente. A pesar de estos bajos valores, las activaciones se registraron en el área Wernicke, circunvolución temporal media y circunvolución d e Heschl. La extensión de las activaciones fue pequeña, 5.06±2.99 cm³, probablemente debido a la baja relación señal ruido del resonador (SNR), además de la complejidad cognitiva de la tarea y el ruido acústico ambiente. Nuestros resultados indican que la implementación de la RMf en comprensión del lenguaje es posible de realizar en Chile sin recursos adicionales.