Search

The surge of supervised learning methods for segmentation lately has underscored the critical role of label quality in predicting performance. This issue is prevalent in the domain of medical imaging, where high annotation costs and inter-observer variability pose significant challenges. Acquiring labels commonly involves multiple experts providing their interpretations of the “true” segmentation labels, each influenced by their individual biases. The blind acceptance of these noisy labels as the ground truth restricts the potential effectiveness of segmentation algorithms. Here, we apply coupled convolutional neural network approaches to a small-sized real-world dataset of bovine cumulus oocyte complexes. This is the first time these methods have been applied to a real-world annotation medical dataset, since they were previously tested only on artificially generated labels of medical and non-medical datasets. This dataset is crucial for healthy embryo development. Its application revealed an important challenge: the inability to effectively learn distinct confusion matrices for each expert due to large areas of agreement. In response, we propose a novel method that focuses on areas of high uncertainty. This approach allows us to understand the individual characteristics better, extract their behavior, and use this insight to create a more sophisticated ground truth using maximum likelihood. These findings contribute to the ongoing discussion of leveraging machine learning algorithms for medical image segmentation, particularly in scenarios involving multiple human annotators.

[EN]Abstract Background: Residual motor deficits of the upper limb in patients with chronic stroke are common and have a negative impact on autonomy, participation and quality of life. Music-Supported Therapy (MST) is an effective intervention to enhance motor and cognitive function, emotional well-being and quality of life in chronic stroke patients. We have adapted the original MST training protocol to a home-based intervention, which incorporates increased training intensity and variability, group sessions, and optimisation of learning to promote autonomy and motivation. Methods: A randomised controlled trial will be conducted to test the effectiveness of this enriched MST (eMST) protocol in improving motor functions, cognition, emotional well-being and quality of life of chronic stroke patients when compared to a program of home-based exercises utilizing the Graded Repetitive Arm Supplementary Program (GRASP). Sixty stroke patients will be recruited and randomly allocated to an eMST group (n = 30) or a control GRASP intervention group (n = 30). Patients will be evaluated before and after a 10-week intervention, as well as at 3-month follow-up. The primary outcome of the study is the functionality of the paretic upper limb measured with the Action Research Arm Test. Secondary outcomes include other motor and cognitive functions, emotional well-being and quality of life measures as well as self-regulation and self-efficacy outcomes. Discussion: We hypothesize that patients treated with eMST will show larger improvements in their motor and cognitive functions, emotional well-being and quality of life than patients treated with a home-based GRASP intervention. Trial registration: The trial has been registered at ClinicalTrials.gov and identified as NCT04507542 on 8 August 2020

Research question Is it possible to identify accurately the optimal first dose of FSH in ovarian stimulation by means of a machine learning model? Design Observational study (2011¿2021) including first IVF cycles with own oocytes. A total of 2713 patients from five private reproductive centres were included in the development phase (2011¿2019) and 774 in the validation phase (2020¿2021). Predictor variables included age, BMI, AMH, AFC and previous live births. Performance was measured with a proposed score based on the number of MII oocytes retrieved and dose received, recommended, or both. Results The included cycles were from women aged 37.7 ± 4.4 years (18¿45 years), with a BMI of 23.5 ± 4.2 kg/m2, AMH of 2.4 ± 2.3 ng/ml, AFC of 11.3 ± 7.6, and an average number of MII obtained 6.9 ± 5.4. The model reached a mean performance score of 0.87 (95% CI 0.86 to 0.88) in the development phase, significantly better than for doses prescribed by clinicians for the same patients (0.83, 95% CI 0.82 to 0.84; P¿=¿2.44 e-10). Mean performance score of the model recommendations was 0.89 (95% CI 0.88 to 0.90) in the validation phase, also significantly better than clinicians (0.84, 95% CI 0.82 to 0.86; P¿=¿3.81 e-05). The model was shown to surpass the performance of standard practice. Conclusion This machine learning model could be used as a training and learning tool for new clinicians, and as quality control for experienced clinicians.

Stability of the menstrual cycle is a key indicator of health, and its alteration can affect the physical, emotional, sexual, and social aspects of menstruating individuals¿ lives. A recently published study showed a statistically significant increase in cycle length after vaccination against COVID-19 and no significant changes in the menses length. However, there is no information about the potential association between vaccination time and change in cycle length. This study aims at assessing the association between the phase of the menstrual cycle at vaccination time and change in cycle length.

[EN]In the Play&Sing project, we are developing an AI platform to support home-based self-training interventions for chronic stroke patients. A large percentage of patients suffering from this disease show motor deficits that clearly hinder their daily activities and diminish their quality of life. In this project we are proposing and testing a new Music Supported Therapy (MST) to induce upper limb motor recovery.With the help of a tablet-based application and a small musical keyboard, we are developing an AI platform to support home-based MST. Specifically, the role of AI algorithms is to support therapists and to boost user engagement by personalizing the interventions according to patient needs and preferences. AI algorithms will provide the therapists with hindsight and foresight tools. In the proposed MST, patients are performing 30 training sessions of 45 minutes with a frequency of 3 sessions per week. In this paper we present our platform and preliminary experiments conducted at a pilot phase.

Case-Based Reasoning (CBR) methodology’s approach to problem-solving that “similar problems have similar solutions” has proved quite favorable for many industrial artificial intelligence applications. However, CBR’s very advantages hinder its performance as case bases (CBs) grow larger than moderate sizes. Searching similar cases is expensive. This handicap often makes CBR less appealing for today’s ubiquitous data environments while, actually, there is ever more reason to benefit from this effective methodology. Accordingly, CBR community’s traditional approach of controlling CB growth to maintain performance is shifting towards finding new ways to deal with abundant data. As a contribution to these efforts, this thesis aims to speed up CBR by leveraging both problem and solution spaces in large-scale CBs that are composed of temporally related cases, as in the example of electronic health records. For the occasions when the speed-up we achieve for exact results may still not be feasible, we endow the CBR system with anytime algorithm capabilities to provide approximate results with confidence upon interruption. Exploiting the temporality of cases allows us to reach superior gains in execution time for CBs of millions of cases. Experiments with publicly available real-world datasets encourage the continued use of CBR in domains where it historically excels like healthcare; and this time, not suffering from, but enjoying big data.

Deciding which are the best treatments is a complex task when patients suffer multiple impairments and when a multidisciplinary team is involved in the intervention. There is always more than a unique treatment option and the results sometimes can be viewed in a short period or only be capable to be measured when the treatment is finished. In this context, the design of effective Clinical Decision Support Systems (CDSS) to help clinicians to select most appropriate interventions is still a challenge. The amount of available data is not always the same for all patients, especially in early treatment stages, hindering the inference in CDSS. To improve the capabilities of CDSS, different components are proposed within a CDSS framework for long-term treatments. A first component is focused on improving the quality of the inferences in missing data scenarios. The Dynamic Multiple Imputation (DMI) algorithm is presented as an effective methodology for data enhancement in CDSS. DMI is capable to adapt to different scenarios with a low or high percentage of missing data. Several experiments conducted reveal that DMI is competitive with regression problems. A second component is devoted to weigh confidence measures, given the uncertainty associated to missing information, by incorporating Mutual Information measures in confidence existing estimators. A third component, based on a community detection algorithm, is proposed to find relationships between clinical decisions that are not explicit. Finally, to illustrate the applicability of different proposed components, two real clinical use cases with chronic patients are presented. The first in the hospital context and the other in the home context.

This work is about speeding up retrieval in Case-Based Reasoning (CBR) for large-scale case bases (CBs) comprised of temporally related cases in metric spaces. A typical example is a CB of electronic health records where consecutive sessions of a patient forms a sequence of related cases. k-Nearest Neighbors (kNN) search is a widely used algorithm in CBR retrieval. However, brute-force kNN is impossible for large CBs. As a contribution to efforts for speeding up kNN search, we introduce an anytime kNN search methodology and algorithm. Anytime Lazy kNN finds exact kNNs when allowed to run to completion with remarkable gain in execution time by avoiding unnecessary neighbor assessments. For applications where the gain in exact kNN search may not suffice, it can be interrupted earlier and it returns best-so-far kNNs together with a confidence value attached to each neighbor. We describe the algorithm and methodology to construct a probabilistic model that we use both to estimate confidence upon interruption and to automatize the interruption at desired confidence thresholds. We present the results of experiments conducted with publicly available datasets. The results show superior gains compared to brute-force search. We reach to an average gain of 87.18% with 0.98 confidence and to 96.84% with 0.70 confidence.

In this paper we present a Virtual Reality game related to Cultural Heritage. We contribute with an analysis of subjective measures taken from questionnaires filled by users after the VR experience, and objective measures taken from logs during the VR game. Specifically, we were interested on study data globally and in groups of user behaviour. Analysing data globally we see a high value of users¿ subjective perceptions. Nevertheless, we found differences of subjective measures when splitting the Novice group. Specifically, the subjective perception of Strugglers is considerably lower than the rest of groups, and this difference is significant. Then, we propose strategies to provide a better experience to Strugglers. We also found correlations between objective and subjective data when they were analysed globally (i.e. without using groups), but these measures did not correlate when they were analysed using behaviour groups.

Archaeological excavations provide us with important clues about the past. Excavated artefacts represent an important connection to civilisations that no longer exist and help us understand some of their customs, traditions and common practices. With the help of academics and practitioners from various disciplines, the results of archaeological excavations can be analysed and a body of knowledge about the corresponding society can be created and shared with members of the general public. Museums have traditionally served the purpose of communicating this knowledge and backing it up with the help of the excavated artefacts. Many museum visitors, however, find it difficult to develop a coherent understanding of the corresponding society only based on the artefacts and annotations shown in museums. Effective modern techniques that have high potential in helping museum visitors with better understanding of the past are 3D reconstruction and virtual reality. 3D reconstruction offers a cost-effective way of recreating historical settlements in a computer-generated virtual environment, while virtual reality helps with immersing people into such environments and reaching a high degree of realism. With the help of these technologies, it becomes possible to relive history, imagine yourself being a part of the reconstructed society and learn about its culture firsthand. The combination of 3D reconstruction and virtual reality represents a very powerful learning tool; however, this tool has been rarely used in a museum setting and its correct use has not been properly investigated. In this paper, we present a study into using virtual reality in itinerant archaeological exhibitions. We discuss the lessons we have learned from developing an interactive virtual reality simulation of the Neolithic settlement of La Draga. These lessons feature our analysis of qualitative and quantitative feedback of museum visitors, as well as what we have learned from analysing their navigation and

Case-Based Reasoning (CBR) is a lazy learning method and, being such, when a new query is made to a CBR system, the swiftness of its retrieval phase proves to be very important for the overall system performance. The availability of ubiquitous data today is an opportunity for CBR systems as it implies more cases to reason with. Nevertheless, this availability also introduces a challenge for the CBR retrieval since distance calculations become computationally expensive. A good example of a domain where the case base is subject to substantial growth over time is the health records of patients where a query is typically an incremental update to prior cases. To deal with the retrieval performance challenge in such domains where cases are sequentially related, we introduce a novel method which significantly reduces the number of cases assessed in the search of exact nearest neighbors (NNs). In particular, when distance measures are metrics, they satisfy the triangle inequality and our method leverages this property to use it as a cutoff in NN search. Specifically, the retrieval is conducted in a lazy manner where only the cases that are true NN candidates for a query are evaluated. We demonstrate how a considerable number of unnecessary distance calculations is avoided in synthetically built domains which exhibit different problem feature characteristics and different cluster diversity.

The aim within this work is to explore the different steps in a data mining process. We will take a methodological point of view, exploring the usual steps in a Knowledge Discovery in Databases process. Perhaps of that focus, the project also wants to give results in a concrete application of those techniques to a real problem. We are collaborating with the Terrassa Hospital physicology department, which give us the data and help us all the way in the interpretation of the results. The database concerns batteries of psychological tests each related with a psychological function. We try to find common patterns or structures inside the database that could give rise to new knowledge related with the diagnosis of Alzheimer in its earlier phases.

This MSc final project focuses on the problem of reconciling the fun aspect of an educational serious video game with its learning objective. For this, we propose bringing video game personalization to educational serious games in order to help players achieve their learning objectives by modifying the game environment dynamically in response to the player¿s behavior. Most applications of video game personalization, however, require detailed models of the player using a lot of information that is often not practical or even possible to get from them. Despite the fact that reinforcement learning in the field of video games has mainly been used for playing, we believe that it is a valuable tool which we will use to achieve an adaptive environment without needing a model of the player. Specifically, we will use the Q-Learning technique to train an AI to help simulated players with simple behaviors fulfill learning objectives in a simplified version of the game. With a virtual reality archaeological educational serious game as our case study, we collected a lot of data and obtained insight into this problem. First, this project studies this data to then try to clearly formalize the problem and decide an appropriate approach. We will also use this data to cluster our players into four types to use for the simulations of different players. Lastly, we turn our attention in a way to evaluate the policies obtained from training and visualize them to better understand what was learned. We will also propose the best course of action when we do not yet know what type of player we are dealing with in a particular playthrough.

In recent years, the Guttmann Institute has incorporated an intelligent assistant as a predicted and personalized decision support system (PPDSS). This PPDSS helps plan rehabilitation sessions for patients suffering from acquired brain injury (ABI). Results show questionable planning when comparing patient profiles and their assigned tasks. The distribution of percentage of effort does not perfectly match the distribution of the cognitive profile. This paper provides a thorough analysis of the patient profiles, showing that a patient’s initial profile and the task execution scores during their first few sessions can be used to better predict their final improvement, to a certain degree of accuracy. Furthermore, results show that more executions of tasks does not automatically lead to improvement. Practice does not seem to make perfect. The proposed technique involves the incorporation of task-weights in the new scheduler.

The interaction between patients and professionals in complex clinical domains, as in the case of Neurorehabilitation, is always a complex process where crucial decision making in a short period of time is required, and where every decision has a serious impact on the patient. In this situation, deciding which are the most appropriate interventions is not an easy task because these patients simultaneously present several impairments, multiple diagnoses, and required complex interdisciplinary approaches. In this context, a methodology and a tool based on ICF have been developed to explore the relationships between patient impairments and therapeutic goals. The proposed approach, based on graph analysis, was used to analyze a set of 1960 patients that suffered an Acquired Brain Injury. Results achieved show that the proposed methodology is able to find non-explicit relationships. This study constitute a first step to the goal of designing a clinical decision support tool for neurorehabilitation.

Introduccion: Changes on electrodermal activity implied variations of skin conductance. It has been proposed as one of the best indicator of changes in sympathetic arousal and emotional states [1]. Its relevance has been explored in several clinical areas such as Epilepsy [3] or Autism Spectrum Disorder [4]. Substance use disorders are a public health concern, not only because its direct damage but also the medical and psychiatric commorbidities. It is not well known about factors that contribute to a better outcomes in detoxification process and abstinence achievement. Substance use disorders could increase arousal and anxiety levels during detoxification and withdrawal process trying to achieve abstinence. These signs and symptoms may affect electrodermal activity. We designed a preliminary study to validate its possible contribution in detoxification processes. © 2017 Elsevier B.V. and ECNP. Published by Elsevier Inc. All rights reserved.

Cash management is concerned with optimizing a company's short-term funding requirements. To this end, various different optimization strategies have been proposed for minimizing costs, using daily cash flow forecasts as the main input to the models. However, the effect of the accuracy of such forecasts on cash management policies has not been studied. This article uses two real data sets from the textile industry to show that the predictive accuracy is highly correlated with cost savings when daily forecasts are used in cash management models. A new method is proposed to help cash managers estimate whether their efforts in improving the predictive accuracy are rewarded by proportional cost savings. Our results indicate the need for an analysis of the potential cost savings derived from improving the predictive accuracy. On that basis, the search for better forecasting models is in place in order to improve cash management.

Finding repeated patterns or motifs in a time series is an important unsupervised task that has still a number of open issues, starting by the definition of motif. In this paper, we revise the notion of motif support, characterizing it as the number of patterns or repetitions that define a motif. We then propose GENMOTIF, a genetic algorithm to discover motifs with support which, at the same time, is flexible enough to accommodate other motif specifications and task characteristics. GENMOTIF is an anytime algorithm that easily adapts to many situations: searching in a range of segment lengths, applying uniform scaling, dealing with multiple dimensions, using different similarity and grouping criteria, etc. GENMOTIF is also parameter-friendly: it has only two intuitive parameters which, if set within reasonable bounds, do not substantially affect its performance. We demonstrate the value of our approach in a number of synthetic and real-world settings, considering traffic volume measurements, accelerometer signals, and telephone call records. © 2016 IEEE.

The detection of very similar patterns in a time series, commonly called motifs, has received continuous and increasing attention from diverse scientific communities. In particular, recent approaches for discovering similar motifs of different lengths have been proposed. In this work, we show that such variable-length similarity-based motifs cannot be directly compared, and hence ranked, by their normalized dissimilarities. Specifically, we find that length-normalized motif dissimilarities still have intrinsic dependencies on the motif length, and that lowest dissimilarities are particularly affected by this dependency. Moreover, we find that such dependencies are generally non-linear and change with the considered data set and dissimilarity measure. Based on these findings, we propose a solution to rank (previously obtained) motifs of different lengths and measure their significance. This solution relies on a compact but accurate model of the dissimilarity space, using a beta distribution with three parameters that depend on the motif length in a non-linear way. We believe the incomparability of variable-length dissimilarities could have an impact beyond the field of time series, and that similar modeling strategies as the one used here could be of help in a more broad context and in diverse application scenarios.

Efficiently finding similar segments or motifs in time series data is a fundamental task that, due to the ubiquity of these data, is present in a wide range of domains and situations. Because of this, countless solutions have been devised but, to date, none of them seems to be fully satisfactory and flexible. In this article, we propose an innovative standpoint and present a solution coming from it: an anytime multimodal optimization algorithm for time series motif discovery based on particle swarms. By considering data from a variety of domains, we show that this solution is extremely competitive when compared to the state-of-the-art, obtaining comparable motifs in considerably less time using minimal memory. In addition, we show that it is robust to different implementation choices and see that it offers an unprecedented degree of flexibility with regard to the task. All these qualities make the presented solution stand out as one of the most prominent candidates for motif discovery in long time series streams. Besides, we believe the proposed standpoint can be exploited in further time series analysis and mining tasks, widening the scope of research and potentially yielding novel effective solutions.

[EN]Being able to predict the performance of a Case-Based Reasoning system against a set of future problems would provide invaluable information for design and maintenance of the system. Thus, we could carry out the needed design changes and maintenance tasks to improve future performance in a proactive fashion. This paper proposes a novel method for identifying regions in a case base where the system gives low confidence solutions to possible future problems. Experimentation is provided for RoboSoccer domain and we argue how encountered regions of dubiosity help us to analyse the case base and the reasoning mechanisms of the given Case-Based Reasoning system.

[EN]When AI technologies are applied to real-world problems, it is often difficult for developers to anticipate all the knowledge needed. Previous research has shown that introspective reasoning can be a useful tool for helping to address this problem in case-based reasoning systems, by enabling them to augment their routine learning of cases with learning to make better use of their cases, as problem-solving experience reveals deficiencies in their reasoning process. In this paper we present a new introspective model for autonomously improving the performance of a CBR system by reasoning about system problem solving failures. We illustrate its benefits with experimental results from tests in an industrial design application.

Coordination infrastructures play a central role in the engineering of multiagent systems. Since the advent of agent technology, research on coordination infrastructures has produced a significant number of infrastructures with varying features. In this paper, we review the the state-of-the-art coordination infrastructures with the purpose of identifying open research challenges that next generation coordination infrastructures should address. Our analysis concludes that next generation coordination infrastructures must address a number of challenges: (i) to become socially aware, by facilitating human interaction within a MAS; (ii) to assist agents in their decision making by providing decision support that helps them reduce the scope of reasoning and facilitates the achievement of their goals; and (iii) to increase openness to support on-line, fully decentralised design and execution. Furthermore, we identify some promising approaches in the literature, together with the research issues worth investigating, to cope with such challenges. © Cambridge University Press, 2015.

Internet-based scenarios, like co-working, e-freelancing, or crowdsourcing, usually need supporting collaboration among several actors that compete to service tasks. Moreover, the distribution of service requests, i.e., the arrival rate, varies over time, as well as the service workload required by each customer. In these scenarios, coalitions can be used to help agents to manage tasks they cannot tackle individually. In this paper we present a model to build and adapt coalitions with the goal of improving the quality and the quantity of tasks completed. The key contribution is a decision making mechanism that uses reputation and adaptation to allow agents in a competitive environment to autonomously enact and sustain coalitions, not only its composition, but also its number, i.e., how many coalitions are necessary. We provide empirical evidence showing that when agents employ our mechanism it is possible for them to maintain high levels of customer satisfaction. First, we show that coalitions keep a high percentage of tasks serviced on time despite a high percentage of unreliable workers. Second, coalitions and agents demonstrate that they successfully adapt to a varying distribution of customers' incoming tasks. This occurs because our decision making mechanism facilitates coalitions to disband when they become non-competitive, and individual agents detect opportunities to start new coalitions in scenarios with high task demand. © 2015 Elsevier Ltd. All rights reserved.

This work presents a method for the extraction of instrumental controls during guitar performances. The method is based on the analysis of multimodal data consisting of a combination of motion capture, audio analysis and musical score. High speed video cameras based on marker identification are used to track the position of finger bones and articulations and audio is recorded with a transducer measuring vibration on the guitar body. The extracted parameters are divided into left hand controls, i.e. fingering (which string and fret is pressed with a left hand finger) and right hand controls, i.e. the plucked string, the plucking finger and the characteristics of the pluck (position, velocity and angles with respect to the string). Controls are estimated based on probability functions of low level features, namely, the plucking instants (i.e. note onsets), the pitch and the distances of the fingers (both hands) to strings and frets. Note onsets are detected via audio analysis, the pitch is extracted from the score and distances are computed from 3D Euclidean Geometry. Results show that by combination of multimodal information, it is possible to estimate such a comprehensive set of control features, with special high performance for the fingering and plucked string estimation. Regarding the plucking finger and the pluck characteristics, their accuracy gets lower but improvements are foreseen including a hand model and the use of high-speed cameras for calibration and evaluation.

Automatically inferring the structural properties of raw multimedia documents is essential in today's digitized society. Given its hierarchical and multi-faceted organization, musical pieces represent a challenge for current computational systems. In this article, we present a novel approach to music structure annotation based on the combination of structure features with time series similarity. Structure features encapsulate both local and global properties of a time series, and allow us to detect boundaries between homogeneous, novel, or repeated segments. Time series similarity is used to identify equivalent segments, corresponding to musically meaningful parts. Extensive tests with a total of five benchmark music collections and seven different human annotations show that the proposed approach is robust to different ground truth choices and parameter settings. Moreover, we see that it outperforms previous approaches evaluated under the same framework. © 1999-2012 IEEE.

In this article we tackle the problem of maximizing cooperation among self-interested agents in a resource exchange environment. Our main concern is the design of mechanisms for maximizing cooperation among self-interested agents in a way that their profits increase by exchanging or trading with resources. Although dynamic coalition formation and partner switching (rewiring) have been shown to promote the emergence and maintenance of cooperation for self-interested agents, no prior work in the literature has investigated whether merging both mechanisms exhibits positive synergies that lead to increase cooperation even further. Therefore, we introduce and analyze a novel dynamic coalition formation mechanism, that uses partner switching, to help self-interested agents to increase their profits in a resource exchange environment. Our experiments show the effectiveness of our mechanism at increasing the agents' profits, as well as the emergence of trading as the preferred behavior over different types of complex networks. © 2014 ACM.

Time series are ubiquitous, and a measure to assess their similarity is a core part of many computational systems. In particular, the similarity measure is the most essential ingredient of time series clustering and classification systems. Because of this importance, countless approaches to estimate time series similarity have been proposed. However, there is a lack of comparative studies using empirical, rigorous, quantitative, and large-scale assessment strategies. In this article, we provide an extensive evaluation of similarity measures for time series classification following the aforementioned principles. We consider 7 different measures coming from alternative measure 'families', and 45 publicly-available time series data sets coming from a wide variety of scientific domains. We focus on out-of-sample classification accuracy, but in-sample accuracies and parameter choices are also discussed. Our work is based on rigorous evaluation methodologies and includes the use of powerful statistical significance tests to derive meaningful conclusions. The obtained results show the equivalence, in terms of accuracy, of a number of measures, but with one single candidate outperforming the rest. Such findings, together with the followed methodology, invite researchers on the field to adopt a more consistent evaluation criteria and a more informed decision regarding the baseline measures to which new developments should be compared. © 2014 Elsevier B.V. All rights reserved.