Search

[EN] The design and control of artificial hands remains a challenge in engineering. Popular prostheses are bio-mechanically simple with restricted manipulation capabilities, as advanced devices are pricy or abandoned due to their difficult communication with the hand. For social robots, the interpretation of human intention is key for their integration in daily life. This can be achieved with machine learning (ML) algorithms, which are barely used for grasping posture recognition. This work proposes an ML approach to recognize nine hand postures, representing 90% of the activities of daily living in real time using an sEMG human-robot interface (HRI). Data from 20 subjects wearing a Myo armband (8 sEMG signals) were gathered from the NinaPro DS5 and from experimental tests with the YCB Object Set, and they were used jointly in the development of a simple multi-layer perceptron in MATLAB, with a global percentage success of 73% using only two features. GPU-based implementations were run to select the best architecture, with generalization capabilities, robustness-versus-electrode shift, low memory expense, and real-time performance. This architecture enables the implementation of grasping posture recognition in low-cost devices, aimed at the development of affordable functional prostheses and HRI for social robots.

Background Excessive alcohol consumption has been associated with increased risk of atrial fibrillation, although the underlying mechanisms remain unclear. An enlarged left atrium and impaired left atrial function may lead to atrial fibrillation. The association of alcohol consumption with structural and functional left atrial measures, however, has received limited attention.Methods and Results We studied 503 participants from the PREDIMED-Plus (Prevencion con Dieta Mediterranea) trial, a randomized trial testing intensive weight loss intervention with an energy-reduced Mediterranean diet and physical activity promotion in preventing cardiovascular disease in adults with metabolic syndrome. Participants underwent transthoracic echocardiography at baseline, year 3, and year 5 of the study. Outcomes of interest included volume index and reservoir, conduit, and contractile strains of the left atrium. Alcohol consumption was calculated through food frequency questionnaires and presented as drinks consumed per day. Multiple linear regression and mixed models estimated the association of alcohol consumption with left atrial measurements at baseline and through follow-up. Cross-sectionally, higher alcohol consumption (per 1 drink/day increases) was associated with larger left atrial volume (0.65 mL/m2 [95% CI, 0.18-1.11]) and lower left atrial reservoir and contractile strain (-0.44% [95% CI, -0.87 to -0.01]; and -0.44% [95% CI, -0.75 to -0.14]). Baseline alcohol consumption was not associated with changes in left atrial measurements, but increases in alcohol consumption (per 1 drink/day increase) during follow-up were associated with left atrial enlargement (0.71 mL/m2 [95% CI, 0.17-1.26]).Conclusions In a population at high cardiovascular risk, increased alcohol consumption

La gravedad de la pandemia producida por la Covid-19 ha promovido el desarrollo de proyectos que buscan soluciones a la crisis sanitaria derivada de esta enfermedad. El Complejo Hospitalario Universitario A Coruña (CHUAC) con el liderazgo del Grupo de Visión Artificial y Reconocimiento de Patrones (VARPA) del INIBIC-UDC han participado en un proyecto nacional con el fin de desarrollar un sistema de diagnóstico temprano de Covid-19 y otras patologías a partir de radiografías de tórax mediante técnicas de inteligencia artificial [1, 2, 3]. La incorporación al sistema sanitario público de sistemas expertos desarrollados en el ámbito académico presenta dos retos principales: El grado de aceptación de una herramienta novedosa por parte de los usuarios y el correcto manejo de información sensible. En este trabajo se muestra como desde el CHUAC se han afrontado estos retos a la hora de incorporar el sistema inteligente a la rutina de los facultativos.

[ES] El cáncer de próstata es a nivel mundial el segundo tipo de cáncer con mayor prevalencia. En 2018 se diagnosticaron 1.3 millones de pacientes y se estima que el número de casos anuales nuevos aumente en un 40.2% en 2030. Esta patología es diagnosticada a partir del análisis visual de biopsias por medio del patólogo y la clasificación de la diferenciación del tejido según la escala Gleason. Esta escala va de 3 a 5, siendo inversamente proporcional al grado de diferenciación. Este proceso diagnóstico es una tarea que consume grandes cantidades de tiempo, y sufre de una elevada variabilidad entre patólogos. Para reducir la carga de trabajo y aumentar el nivel de objetividad, en los últimos años se han propuesto sistemas de ayuda al diagnóstico basados en algoritmos de deep learning y, en concreto, redes neuronales convoluciones. A pesar de los prometedores resultados obtenidos por estos sistemas, las técnicas punteras aún se ven limitadas, entre otros factores, por el sesgo introducido en la anotación durante el entrenamiento. Por ello, el objetivo de este TFG es el desarrollo de modelos de deep learning robustos a la variabilidad entre patólogos durante el entrenamiento con tal de mejorar la capacidad de generalización. En esta línea, se pretende utilizar funciones de pérdida basadas en el estadístico cuadrático de Cohen, y en suavizado de etiquetas. Asimismo, se tratará de optimizar las arquitecturas de redes neuronales convolucionales utilizadas en el estado del arte, incluyendo bloques de atención. Con tal de validar la mejora en la capacidad de generalización, estos métodos serán entrenados y testeados en distintas bases de datos, con etiquetas de referencia dadas por distintos patólogos., [EN] Prostate cancer is the second most prevalent type of cancer worldwide. In 2018, 1.3 million patients were diagnosed and it is estimated that the number of new annual cases will increase by 40.2% in 2030. This pathology is diagnosed from the visual analysis of biopsies by pathologists and the later classification of tissue differentiation according to the Gleason scale. This scale goes from 3 to 5, being inversely proportional to the degree of differentiation. This diagnostic process is a timeconsuming task, and suffers from high variability among pathologists. To reduce the workload and increase the level of objectivity, in recent years, diagnostic support systems have been proposed based on deep learning algorithms and, specifically, convolutional neural networks. Despite the promising results obtained by these systems, state­of­the­art techniques are still limited, among other factors, by the bias introduced into the annotation during training. Therefore, the aim of this thesis is the development of deep learning models that are robust to the variability between pathologists during training, in order to improve their generalization capacity. Following this purpose, loss functions based on Cohen¿s quadratic statistic will be used, as well as label smoothing. Likewise, this thesis will investigate the possibilities of optimizing convolutional neural network architectures that are used in the state of the art, including attention blocks which include multi­level features into the decision. In order to validate the improvement in generalization capacity, these methods will be trained and tested in different databases, with reference labels given by different pathologists. Finally, the results obtained will be analyzed, these have shown that providing the neural network with information on inter­pathologist variability improves generalization to external databases.

To determine the risk of mortality and need for aortic valve replacement (AVR) in patients with low-flow low-gradient (LFLG) aortic stenosis (AS). A longitudinal multicentre study including consecutive patients with severe AS (aortic valve area [AVA] < 1.0 cm 2) and normal left ventricular ejection fraction (LVEF). Patients were classified as: high-gradient (HG, mean gradient ≥ 40 mmHg), normal-flow low-gradient (NFLG, mean gradient < 40 mmHg, indexed systolic volume (SVi) > 35 ml/m 2) and LFLG (mean gradient < 40 mmHg, SVi ≤ 35 ml/m 2). Of 1,391 patients, 147 (10.5%) had LFLG, 752 (54.1%) HG, and 492 (35.4%) NFLG. Echocardiographic parameters of the LFLG group showed similar AVA to the HG group but with less severity in the dimensionless index, calcification, and hypertrophy. The HG group required AVR earlier than NFLG (p < 0.001) and LFLG (p < 0.001), with no differences between LFLG and NFLG groups (p = 0.358). Overall mortality was 27.7% (CI 95% 25.3-30.1) with no differences among groups (p = 0.319). The impact of AVR in terms of overall mortality reduction was observed the most in patients with HG (hazard ratio [ HR ]: 0.17; 95% CI : 0.12-0.23; p < 0.001), followed by patients with LFLG (HR : 0.25; 95% CI : 0.13-0.49; p < 0.001), and finally patients with NFLG (HR : 0.29; 95% CI : 0.20-0.44; p < 0.001), with a risk reduction of 84, 75, and 71%, respectively. Paradoxical LFLG AS affects 10.5% of severe AS, and has a lower need for AVR than the HG group and similar to the NFLG group, with no differences in mortality. AVR had a lower impact on LFLG AS compared with HG AS. Therefore, the findings of the present study showed LFLG AS to have an intermediate clinical risk profile betwee the HG and NFHG groups

“Early in 2021 we were eager on what to expect in year two of the pandemic. The extraordinary progress and success of vaccination, saving several million human lives in the most developed countries, worldwide projected “vaccine” (and therefore science) as the word of the year. Smart, open, and innovative approaches were developed within the CSIC Global Health PTI (Interdisciplinary Thematic Platforms) in a creative and multidisciplinary atmosphere, and our Institute successfully participated in the cross-cutting working group on data and artificial intelligence applications. We were also deeply involved in three of the CSIC2030 “white papers” on Global Change Impacts, Oceans, and Brain and Behaviour. Despite the difficulties, CEAB has produced more science than ever in this period and it is now a well routed Institute worldwide recognized in Ecology, Marine Ecology and Aquatic Sciences, with several relevant publications in the top journals Biological Conservation, Ecology, Frontiers in Marine Science, Global Change Biology, Global Biogeochemical Cycles, Science of the Total Environment, Science Advances, PNAS, and Nature Ecology&Evolution, among others. Congratulations to the big CEAB family, scientists, technicians and administrative staff, all of them necessary for producing the best standards in science. Finally, I want to add a few personal words. I am coming to the end of my period of Director of CEAB in late 2022. It has been a big pleasure, and a very stimulating journey plenty of challenges to be leading the transformative refoundation of CEAB for the last 8 yrs. I enjoyed it very much and I send my best wishes for the new Director and the new dreams CEAB may reach. If you can imagine it, it can be done.