Abstract

Abstract
Accurate estimation of hand shape and position is an important task in various applications, such as human-computer interaction, human-robot interaction, and virtual and augmented reality. In this paper, it is proposed a method to estimate the hand keypoints from single and colored images utilizing the pre-trained deep convolutional neural networks VGG-16 and VGG-19. The method is evaluated on the FreiHAND dataset, and the performance of the two neural networks is compared. The best results were achieved by the VGG-19, with average estimation errors of 7.40 pixels and 11.36 millimeters for the best cases of two-dimensional and three-dimensional hand keypoints estimation, respectively.

Abstract
Machine Learning has been overtaken by a growing necessity to explain and understand decisions made by trained models as regulation and consumer awareness have increased. Alongside understanding the inner workings of a model comes the task of verifying how adequately we can model a problem with the learned functions. Traditional global assessment functions lack the granularity required to understand local differences in performance in different regions of the feature space, where the model can have problems adapting. Residual Analysis adds a layer of model understanding by interpreting prediction residuals in an exploratory manner. However, this task can be unfeasible for high-dimensionality datasets through hypotheses and visualizations alone. In this work, we use weak interpretable learners to identify regions of high prediction error in the feature space. We achieve this by examining the absolute residuals of predictions made by trained regressors. This methodology retains the interpretability of the identified regions. It allows practitioners to have tools to formulate hypotheses surrounding model failure on particular regions for future model tunning, data collection, or data augmentation on critical cohorts of data. We present a way of including information on different levels of model uncertainty in the feature space through the use of locally fitted Model Agnostic Prediction Intervals (MAPIE) in the identified regions, comparing this approach with other common forms of conformal predictions which do not take into account findings from weak segment identification, by assessing local and global coverage of the prediction intervals. To demonstrate the practical application of our approach, we present a real-world industry use case in the context of inbound retention call-centre operations for a Telecom Provider to determine optimal pairing between a customer and an available assistant through the prediction of contracted revenue. © 2023, The Author(s), under exclusive license to Springer Nature Switzerland AG.

Abstract
Introduction and Objectives: Obstructive sleep apnea (OSA) is a prevalent sleep condition which is very heterogeneous although not formally characterized as such, resulting in missed or delayed diagnosis. Cluster analysis has been used in different clinical domains, particularly within sleep disorders. We aim to understand OSA heterogeneity and provide a variety of cluster visualizations to communicate the information clearly and efficiently.Materials and Methods: We applied an extension of k-means to be used in categorical variables: k -modes, to identify OSA patients' groups, based on demographic, physical examination, clinical his-tory, and comorbidities characterization variables (n = 40) obtained from a derivation and validation cohorts (211 and 53, respectively) from the northern region of Portugal. Missing values were imputed with k-nearest neighbours (k-NN) and a chi-square test was held for feature selection.Results: Thirteen variables were inserted in phenotypes, resulting in the following three clus-ters: Cluster 1, middle-aged males reporting witnessed apneas and high alcohol consumption before sleep; Cluster 2, middle-aged women with increased neck circumference (NC), non -repairing sleep and morning headaches; and Cluster 3, obese elderly males with increased NC, witnessed apneas and alcohol consumption. Patients from the validation cohort assigned to dif-ferent clusters showed similar proportions when compared with the derivation cohort, for mild (C1: 56 vs 75%, P = 0.230; C2: 61 vs 75%, P = 0.128; C3: 45 vs 48%, P = 0.831), moderate (C1: 24 vs 25%; C2: 20 vs 25%; C3: 25 vs 19%) and severe (C1: 20 vs 0%; C2: 18 vs 0%; C3: 29 vs 33%) levels. Therefore, the allocation supported the validation of the obtained clusters.Conclusions: Our findings suggest different OSA patients' groups, creating the need to rethink these patients' stereotypical baseline characteristics.(c) 2021 Sociedade Portuguesa de Pneumologia. Published by Elsevier Espana, S.L.U. This is an open access article under the CC BY-NC-ND license (http://creativecommons.org/licenses/by-nc-nd/4.0/).

Abstract
[No abstract available]

Abstract
Background: Digital clinical tools are a new technology that can be used in the screening or diagnosis of obstructive sleep apnea (OSA), notwithstanding the crucial role of polysomnography, the gold standard.Objective: This study aimed to identify, gather, and analyze the most accurate digital tools and smartphone-based health platforms used for OSA screening or diagnosis in the adult population. Methods: We performed a comprehensive literature search of PubMed, Scopus, and Web of Science databases for studies evaluating the validity of digital tools in OSA screening or diagnosis until November 2022. The risk of bias was assessed using the Joanna Briggs Institute critical appraisal tool for diagnostic test accuracy studies. The sensitivity, specificity, and area under the curve (AUC) were used as discrimination measures.Results: We retrieved 1714 articles, 41 (2.39%) of which were included in the study. From these 41 articles, we found 7 (17%) smartphone-based tools, 10 (24%) wearables, 11 (27%) bed or mattress sensors, 5 (12%) nasal airflow devices, and 8 (20%) other sensors that did not fit the previous categories. Only 8 (20%) of the 41 studies performed external validation of the developed tool. Of these, the highest reported values for AUC, sensitivity, and specificity were 0.99, 96%, and 92%, respectively, for a clinical cutoff of apnea-hypopnea index (AHI)& GE;30. These values correspond to a noncontact audio recorder that records sleep sounds, which are then analyzed by a deep learning technique that automatically detects sleep apnea events, calculates the AHI, and identifies OSA. Looking at the studies that only internally validated their models, the work that reported the highest accuracy measures showed AUC, sensitivity, and specificity values of 1.00, 100%, and 96%, respectively, for a clinical cutoff AHI & GE;30. It uses the Sonomat-a foam mattress that, aside from recording breath sounds, has pressure sensors that generate voltage when deformed, thus detecting respiratory movements, and uses it to classify OSA events.Conclusions: These clinical tools presented promising results with high discrimination measures (best results reached AUC>0.99). However, there is still a need for quality studies comparing the developed tools with the gold standard and validating them in external populations and other environments before they can be used in clinical settings.