Identifying prognostic factors for survival in intensive care unit patients with SIRS or sepsis by machine learning analysis on electronic health records.

BACKGROUND: Systemic inflammatory response syndrome (SIRS) and sepsis are the most common causes of in-hospital death. However, the characteristics associated with the improvement in the patient conditions during the ICU stay were not fully elucidated for each population as well as the possible differences between the two. GOAL: The aim of this study is to highlight the differences between the prognostic clinical features for the survival of patients diagnosed with SIRS and those of patients diagnosed with sepsis by using a multi-variable predictive modeling approach with a reduced set of easily available measurements collected at the admission to the intensive care unit (ICU). METHODS: Data were collected from 1,257 patients (816 non-sepsis SIRS and 441 sepsis) admitted to the ICU. We compared the performance of five machine learning models in predicting patient survival. Matthews correlation coefficient (MCC) was used to evaluate model performances and feature importance, and by applying Monte Carlo stratified Cross-Validation. RESULTS: Extreme Gradient Boosting (MCC = 0.489) and Logistic Regression (MCC = 0.533) achieved the highest results for SIRS and sepsis cohorts, respectively. In order of importance, APACHE II, mean platelet volume (MPV), eosinophil counts (EoC), and C-reactive protein (CRP) showed higher importance for predicting sepsis patient survival, whereas, SOFA, APACHE II, platelet counts (PLTC), and CRP obtained higher importance in the SIRS cohort. CONCLUSION: By using complete blood count parameters as predictors of ICU patient survival, machine learning models can accurately predict the survival of SIRS and sepsis ICU patients. Interestingly, feature importance highlights the role of CRP and APACHE II in both SIRS and sepsis populations. In addition, MPV and EoC are shown to be important features for the sepsis population only, whereas SOFA and PLTC have higher importance for SIRS patients.

Comparative Assessment of Physiological Responses to Emotional Elicitation by Auditory and Visual Stimuli.

The study of emotions through the analysis of the induced physiological responses gained increasing interest in the past decades. Emotion-related studies usually employ films or video clips, but these stimuli do not give the possibility to properly separate and assess the emotional content provided by sight or hearing in terms of physiological responses. In this study we have devised an experimental protocol to elicit emotions by using, separately and jointly, pictures and sounds from the widely used International Affective Pictures System and International Affective Digital Sounds databases. We processed galvanic skin response, electrocardiogram, blood volume pulse, pupillary signal and electroencephalogram from 21 subjects to extract both autonomic and central nervous system indices to assess physiological responses in relation to three types of stimulation: auditory, visual, and auditory/visual. Results show a higher galvanic skin response to sounds compared to images. Electrocardiogram and blood volume pulse show different trends between auditory and visual stimuli. The electroencephalographic signal reveals a greater attention paid by the subjects when listening to sounds compared to watching images. In conclusion, these results suggest that emotional responses increase during auditory stimulation at both central and peripheral levels, demonstrating the importance of sounds for emotion recognition experiments and also opening the possibility toward the extension of auditory stimuli in other fields of psychophysiology. Clinical and Translational Impact Statement- These findings corroborate auditory stimuli's importance in eliciting emotions, supporting their use in studying affective responses, e.g., mood disorder diagnosis, human-machine interaction, and emotional perception in pathology.

A Screening Platform for Hearing Loss and Cognitive Decline: WHISPER (Widespread Hearing Impairment Screening and PrEvention of Risk).

The WHISPER (Widespread Hearing Impairment Screening and PrEvention of Risk) platform was recently developed for screening for hearing loss (HL) and cognitive decline in adults. It includes a battery of tests (a risk factors (RF) questionnaire, a language-independent speech-in-noise test, and cognitive tests) and provides a pass/fail outcome based on the analysis of several features. Earlier studies demonstrated high accuracy of the speech-in-noise test for predicting HL in 350 participants. In this study, preliminary results from the RF questionnaire (137 participants) and from the visual digit span test (DST) (78 participants) are presented. Despite the relatively small sample size, these findings indicate that the RF and DST may provide additional features that could be useful to characterize the overall individual profile, providing additional knowledge related to short-term memory performance and overall risk of HL and cognitive decline. Future research is needed to expand number of subjects tested, number of features analyzed, and the range of algorithms (including supervised and unsupervised machine learning) used to identify novel measures able to predict the individual hearing and cognitive abilities, also including components related to the individual risk.

Multi-Input Multi-Output Dynamic Modelling of Type 2 Diabetes Progression.

Type 2 Diabetes Mellitus (T2D) is a chronic health condition that affects millions of people globally. Early identification of risk can support preventive intervention and therefore slow down disease progression. Risk characterization is also necessary to monitor the mechanisms behind the pathology through the analysis of the interrelationships between the predictors and their time course. In this work, a multi-input multi-output Gaussian Process model is proposed to describe the evolution of different biomarkers in patients who will/will not develop T2D considering the interdependencies between outputs. The preliminary results obtained suggest that the trends in biomarkers captured by the model are coherent with the literature and with real-world data, demonstrating the value of multi-input multi-output approaches. In future developments, the proposed method could be applied to assess how the biomarkers evolve and interact with each other in groups of patients having in common one or more risk factors.

Characterization of Inclination Analysis for Predicting Onset of Heart Failure from Primary Care Electronic Medical Records.

The aim of this study is to characterize the performance of an inclination analysis for predicting the onset of heart failure (HF) from routinely collected clinical biomarkers extracted from primary care electronic medical records. A balanced dataset of 698 patients (with/without HF), including a minimum of five longitudinal measures of nine biomarkers (body mass index, diastolic and systolic blood pressure, fasting glucose, glycated hemoglobin, low-density and high-density lipoproteins, total cholesterol, and triglycerides) is used. The proposed algorithm achieves an accuracy of 0.89 (sensitivity of 0.89, specificity of 0.90) to predict the inclination of biomarkers (i.e., their trend towards a 'survival' or 'collapse' as defined by an inclination analysis) on a labeled, balanced dataset of 40 patients. Decision trees trained on the predicted inclination of biomarkers have significantly higher recall (0.69 vs. 0.53) and significantly higher negative predictive value (0.60 vs. 0.55) than those trained on the average values computed from the measures of biomarkers available before the onset of the disease, suggesting that an inclination analysis can help identify the onset of HF in the primary care patient population from routinely available clinical data. This exploratory study provides the basis for further investigations of inclination analyses to identify at-risk patients and generate preventive measures (i.e., personalized recommendations to reverse the trend of biomarkers towards collapse).

Characterization of Synthetic Health Data Using Rule-Based Artificial Intelligence Models.

The aim of this study is to apply and characterize eXplainable AI (XAI) to assess the quality of synthetic health data generated using a data augmentation algorithm. In this exploratory study, several synthetic datasets are generated using various configurations of a conditional Generative Adversarial Network (GAN) from a set of 156 observations related to adult hearing screening. A rule-based native XAI algorithm, the Logic Learning Machine, is used in combination with conventional utility metrics. The classification performance in different conditions is assessed: models trained and tested on synthetic data, models trained on synthetic data and tested on real data, and models trained on real data and tested on synthetic data. The rules extracted from real and synthetic data are then compared using a rule similarity metric. The results indicate that XAI may be used to assess the quality of synthetic data by (i) the analysis of classification performance and (ii) the analysis of the rules extracted on real and synthetic data (number, covering, structure, cut-off values, and similarity). These results suggest that XAI can be used in an original way to assess synthetic health data and extract knowledge about the mechanisms underlying the generated data.

Dual-View Single-Shot Multibox Detector at Urban Intersections: Settings and Performance Evaluation.

The explosion of artificial intelligence methods has paved the way for more sophisticated smart mobility solutions. In this work, we present a multi-camera video content analysis (VCA) system that exploits a single-shot multibox detector (SSD) network to detect vehicles, riders, and pedestrians and triggers alerts to drivers of public transportation vehicles approaching the surveilled area. The evaluation of the VCA system will address both detection and alert generation performance by combining visual and quantitative approaches. Starting from a SSD model trained for a single camera, we added a second one, under a different field of view (FOV) to improve the accuracy and reliability of the system. Due to real-time constraints, the complexity of the VCA system must be limited, thus calling for a simple multi-view fusion method. According to the experimental test-bed, the use of two cameras achieves a better balance between precision (68%) and recall (84%) with respect to the use of a single camera (i.e., 62% precision and 86% recall). In addition, a system evaluation in temporal terms is provided, showing that missed alerts (false negatives) and wrong alerts (false positives) are typically transitory events. Therefore, adding spatial and temporal redundancy increases the overall reliability of the VCA system.

A novel method to derive personalized minimum viable recommendations for type 2 diabetes prevention based on counterfactual explanations.

Despite the growing availability of artificial intelligence models for predicting type 2 diabetes, there is still a lack of personalized approaches to quantify minimum viable changes in biomarkers that may help reduce the individual risk of developing disease. The aim of this article is to develop a new method, based on counterfactual explanations, to generate personalized recommendations to reduce the one-year risk of type 2 diabetes. Ten routinely collected biomarkers extracted from Electronic Medical Records of 2791 patients at low risk and 2791 patients at high risk of type 2 diabetes were analyzed. Two regions characterizing the two classes of patients were estimated using a Support Vector Data Description classifier. Counterfactual explanations (i.e., minimal changes in input features able to change the risk class) were generated for patients at high risk and evaluated using performance metrics (availability, validity, actionability, similarity, and discriminative power) and a qualitative survey administered to seven expert clinicians. Results showed that, on average, the requested minimum viable changes implied a significant reduction of fasting blood sugar, systolic blood pressure, and triglycerides and a significant increase of high-density lipoprotein in patients at risk of diabetes. A significant reduction in body mass index was also recommended in most of the patients at risk, except in females without hypertension. In general, greater changes were recommended in hypertensive patients compared to non-hypertensive ones. The experts were overall satisfied with the proposed approach although in some cases the proposed recommendations were deemed insufficient to reduce the risk in a clinically meaningful way. Future research will focus on a larger set of biomarkers and different comorbidities, also incorporating clinical guidelines whenever possible. Development of additional mathematical and clinical validation approaches will also be of paramount importance.

Analysis of the skin conductance and pupil signals for evaluation of emotional elicitation by images and sounds.

Many studies in the literature attempt recognition of emotions through the use of videos or images, but very few have explored the role that sounds have in evoking emotions. In this study we have devised an experimental protocol for elicitation of emotions by using, separately and jointly, images and sounds from the widely used International Affective Pictures System and International Affective Digital Sounds databases. During the experiments we have recorded the skin conductance and pupillary signals and processed them with the goal of extracting indices linked to the autonomic nervous system, thus revealing specific patterns of behavior depending on the different stimulation modalities. Our results show that skin conductance helps discriminate emotions along the arousal dimension, whereas features derived from the pupillary signal are able to discriminate different states along both valence and arousal dimensions. In particular, the pupillary diameter was found to be significantly greater at increasing arousal and during elicitation of negative emotions in the phases of viewing images and images with sounds. In the sound-only phase, on the other hand, the power calculated in the high and very high frequency bands of the pupillary diameter were significantly greater at higher valence (valence ratings > 5). Clinical relevance- This study demonstrates the ability of physiological signals to assess specific emotional states by providing different activation patterns depending on the stimulation through images, sounds and images with sounds. The approach has high clinical relevance as it could be extended to evaluate mood disorders (e.g. depression, bipolar disorders, or just stress), or to use physiological patterns found for sounds in order to study whether hearing aids can lead to increased emotional perception.

Evaluation of Machine Learning Algorithms and Explainability Techniques to Detect Hearing Loss From a Speech-in-Noise Screening Test.

PURPOSE: The aim of this study was to analyze the performance of multivariate machine learning (ML) models applied to a speech-in-noise hearing screening test and investigate the contribution of the measured features toward hearing loss detection using explainability techniques. METHOD: Seven different ML techniques, including transparent (i.e., decision tree and logistic regression) and opaque (e.g., random forest) models, were trained and evaluated on a data set including 215 tested ears (99 with hearing loss of mild degree or higher and 116 with no hearing loss). Post hoc explainability techniques were applied to highlight the role of each feature in predicting hearing loss. RESULTS: Random forest (accuracy = .85, sensitivity = .86, specificity = .85, precision = .84) performed, on average, better than decision tree (accuracy = .82, sensitivity = .84, specificity = .80, precision = .79). Support vector machine, logistic regression, and gradient boosting had similar performance as random forest. According to post hoc explainability analysis on models generated using random forest, the features with the highest relevance in predicting hearing loss were age, number and percentage of correct responses, and average reaction time, whereas the total test time had the lowest relevance. CONCLUSIONS: This study demonstrates that a multivariate approach can help detect hearing loss with satisfactory performance. Further research on a bigger sample and using more complex ML algorithms and explainability techniques is needed to fully investigate the role of input features (including additional features such as risk factors and individual responses to low-/high-frequency stimuli) in predicting hearing loss.

A Novel Approach for Segment-Length Selection Based on Stationarity to Perform Effective Connectivity Analysis Applied to Resting-State EEG Signals.

Connectivity among different areas within the brain is a topic that has been notably studied in the last decade. In particular, EEG-derived measures of effective connectivity examine the directionalities and the exerted influences raised from the interactions among neural sources that are masked out on EEG signals. This is usually performed by fitting multivariate autoregressive models that rely on the stationarity that is assumed to be maintained over shorter bits of the signals. However, despite being a central condition, the selection process of a segment length that guarantees stationary conditions has not been systematically addressed within the effective connectivity framework, and thus, plenty of works consider different window sizes and provide a diversity of connectivity results. In this study, a segment-size-selection procedure based on fourth-order statistics is proposed to make an informed decision on the appropriate window size that guarantees stationarity both in temporal and spatial terms. Specifically, kurtosis is estimated as a function of the window size and used to measure stationarity. A search algorithm is implemented to find the segments with similar stationary properties while maximizing the number of channels that exhibit the same properties and grouping them accordingly. This approach is tested on EEG signals recorded from six healthy subjects during resting-state conditions, and the results obtained from the proposed method are compared to those obtained using the classical approach for mapping effective connectivity. The results show that the proposed method highlights the influence that arises in the Default Mode Network circuit by selecting a window of 4 s, which provides, overall, the most uniform stationary properties across channels.

Characterization of Type 2 Diabetes Using Counterfactuals and Explainable AI.

Type 2 diabetes mellitus is a metabolic disorder of glucose management, whose prevalence is increasing inexorably worldwide. Adherence to therapies, along with a healthy lifestyle can help prevent the onset of disease. This preliminary study proposes the use of explainable artificial intelligence techniques with the aim of (i) characterizing diabetic patients through a set of easily interpretable rules and (ii) providing individualized recommendations for the prevention of the onset of the disease through the generation of counterfactual explanations, based on minimal variations of biomarkers routinely collected in primary care. The results of this preliminary study parallel findings from the literature as differences in biomarkers between patients with and without diabetes are observed for fasting blood sugar, body mass index, and high-density lipoprotein levels.

Behavioral Segmentation for Enhanced Peer-to-Peer Patient Education.

The aim of this study was to develop a peer-to-peer virtual intervention for patients with type 2 diabetes from different segments: patients who take several medications (medication group), patients who do not take diabetes medications (lifestyle group), and a mixed group. Preliminary results showed that patients in the lifestyle group were interested in preventive strategies, reporting better learning experience and higher motivation than those in the medication group. Future research is needed to design approaches tailored to patients in the medication group.

Design for a Virtual Peer-to-Peer Knowledge to Action Platform for Type 2 Diabetes.

Many patients with Type 2 Diabetes (T2D) have difficulty in controlling their disease despite wide-spread availability of high-quality guidelines, T2D education programs and primary care follow-up programs. Current diabetes education and treatment programs translate knowledge from bench to bedside well, but underperform on the 'last-mile' of converting that knowledge into action (KTA). Two innovations to the last-mile problem in management of patients with T2D are introduced. 1) Design of a platform for peer-to-peer groups where patients can solve KTA problems together in a structured and psychologically safe environment using all the elements of the Action Cycle phase of the KTA framework. The platform uses Self-Determination Theory as the behavior change theory. 2) A novel patient segmentation method to enable the formation of groups of patients who have similar behavioral characteristics and therefore who are more likely to find common cause in the fight against diabetes.

Improving Shared Decision-Making Using Cognitive Effort-Optimization.

Diabetes Prevention Programs (DPPs) can prevent or delay type 2 diabetes (T2D). However, the participation rates in DPPs have been limited. Many individuals at risk of developing diabetes have difficulties making healthy choices because of the cognitive effort required to understand the risks, the role of biomarkers, the consequences of inaction and the actions required to delay or avoid development of T2D. We report on the design and development of a prototype digital tool that decreases cognitive effort for people at risk of developing T2D using the effort-optimized intervention framework.

Emotion recognition from multimodal physiological measurements based on an interpretable feature selection method.

Many studies in literature successfully use classification algorithms to classify emotions by means of physiological signals. However, there are still important limitations in interpretability of the results, i.e. lack of feature specific characterizations for each emotional state. To this extent, our study proposes a feature selection method that allows to determine the most informative subset of features extracted from physiological signals by maintaining their original dimensional space. Results show that features from the galvanic skin response are confirmed to be relevant in separating the arousal dimension, especially fear from happiness and relaxation. Furthermore, the average and the median value of the galvanic skin response signal together with the ratio between SD1 and SD2 from the Poincarè analysis of the electrocardiogram signal, were found to be the most important features for the discrimination along the valence dimension. A Linear Discriminant Analysis model using the first ten features sorted by importance, as defined by their ability to discriminate emotions with a bivariate approach, led to a three-class test accuracy in discriminating happiness, relaxation and fear equal to 72%, 67% and 89% respectively.Clinical relevance This study demonstrates the ability of physiological signals to assess the emotional state of different subjects, by providing a fast and efficient method to select most important indexes from the autonomic nervous system. The approach has high clinical relevance as it could be extended to assess other emotional states (e.g. stress and pain) characterizing pathological states such as post traumatic stress disorder and depression.

Evaluation of a Novel Speech-in-Noise Test for Hearing Screening: Classification Performance and Transducers' Characteristics.

One of the current gaps in teleaudiology is the lack of methods for adult hearing screening viable for use in individuals of unknown language and in varying environments. We have developed a novel automated speech-in-noise test that uses stimuli viable for use in non-native listeners. The test reliability has been demonstrated in laboratory settings and in uncontrolled environmental noise settings in previous studies. The aim of this study was: (i) to evaluate the ability of the test to identify hearing loss using multivariate logistic regression classifiers in a population of 148 unscreened adults and (ii) to evaluate the ear-level sound pressure levels generated by different earphones and headphones as a function of the test volume. The multivariate classifiers had sensitivity equal to 0.79 and specificity equal to 0.79 using both the full set of features extracted from the test as well as a subset of three features (speech recognition threshold, age, and number of correct responses). The analysis of the ear-level sound pressure levels showed substantial variability across transducer types and models, with earphones levels being up to 22 dB lower than those of headphones. Overall, these results suggest that the proposed approach might be viable for hearing screening in varying environments if an option to self-adjust the test volume is included and if headphones are used. Future research is needed to assess the viability of the test for screening at a distance, for example by addressing the influence of user interface, device, and settings, on a large sample of subjects with varying hearing loss.

A Generative Adversarial Network (GAN) Technique for Internet of Medical Things Data.

The application of machine learning and artificial intelligence techniques in the medical world is growing, with a range of purposes: from the identification and prediction of possible diseases to patient monitoring and clinical decision support systems. Furthermore, the widespread use of remote monitoring medical devices, under the umbrella of the "Internet of Medical Things" (IoMT), has simplified the retrieval of patient information as they allow continuous monitoring and direct access to data by healthcare providers. However, due to possible issues in real-world settings, such as loss of connectivity, irregular use, misuse, or poor adherence to a monitoring program, the data collected might not be sufficient to implement accurate algorithms. For this reason, data augmentation techniques can be used to create synthetic datasets sufficiently large to train machine learning models. In this work, we apply the concept of generative adversarial networks (GANs) to perform a data augmentation from patient data obtained through IoMT sensors for Chronic Obstructive Pulmonary Disease (COPD) monitoring. We also apply an explainable AI algorithm to demonstrate the accuracy of the synthetic data by comparing it to the real data recorded by the sensors. The results obtained demonstrate how synthetic datasets created through a well-structured GAN are comparable with a real dataset, as validated by a novel approach based on machine learning.

eHealth for patients with rare diseases: the eHealth Working Group of the European Reference Network on Rare Multisystemic Vascular Diseases (VASCERN).

BACKGROUND: The European Reference Network on Rare Multisystemic Vascular Diseases (VASCERN) was launched in 2017 and involves, to date, 35 highly specialised multidisciplinary expert centres (from the 30 full Healthcare Provider members) coming from 11 countries and more than 70 patient organizations from 16 countries. The eHealth Working Group (WG) of VASCERN was set up to develop practical, patient-centred solutions and strategies for effective use of eHealth tools to answer the needs of patients with multisystemic vascular rare diseases. THE EHEALTH WG: Following the identified patients' needs and following the guiding principles of collaboration and patient-centredness, the eHealth WG was created with the following aims: to develop a mobile app to help patients find expert centres and patient organizations, and to develop resources (Pills of Knowledge, PoK) for training and education via digital platforms (eLearning). The mobile app includes, to date, functionalities that allow users to find expert centres and patient organizations across Europe in the area of rare multisystemic vascular diseases. Discussed app developments include personalized digital patient passports, educational material, emergency management guidelines and remote consultations. Regarding training and education, a variety of PoK have been developed. The PoK cover several topics, target several user groups, and are delivered in various formats so that they are easy-to-use, easy-to-understand, informative, and viable for delivery and sharing through digital platforms (eLearning) including, e.g., the VASCERN YouTube™ channel. CONCLUSION: Overall, the work carried out by the eHealth WG of VASCERN can be seen as a pilot experience that may serve as a basis to for collaborative development of patient-centred eHealth tools that answer the needs of patients with various rare diseases, not limited to rare multisystemic vascular diseases. By expanding the multidisciplinary approach here described, clinical and research networks can take advantage of eHealth services and use them as strategic assets in achieving the ultimate goal of ensuring equity of access to prevention programs, timely and accurate diagnosis and specialized care for patients with rare diseases throughout Europe.

Automated classification of cancer morphology from Italian pathology reports using Natural Language Processing techniques: A rule-based approach.

Pathology reports represent a primary source of information for cancer registries. Hospitals routinely process high volumes of free-text reports, a valuable source of information regarding cancer diagnosis for improving clinical care and supporting research. Information extraction and coding of textual unstructured data is typically a manual, labour-intensive process. There is a need to develop automated approaches to extract meaningful information from such texts in a reliable and accurate way. In this scenario, Natural Language Processing (NLP) algorithms offer a unique opportunity to automatically encode the unstructured reports into structured data, thus representing a potential powerful alternative to expensive manual processing. However, notwithstanding the increasing interest in this area, there is still limited availability of NLP approaches for pathology reports in languages other than English, including Italian, to date. The aim of our work was to develop an automated algorithm based on NLP techniques, able to identify and classify the morphological content of pathology reports in the Italian language with micro-averaged performance scores higher than 95%. Specifically, a novel, domain-specific classifier that uses linguistic rules was developed and tested on 27,239 pathology reports from a single Italian oncological centre, following the International Classification of Diseases for Oncology morphology classification standard (ICD-O-M). The proposed classification algorithm achieved successful results with a micro-F1 score of 98.14% on 9594 pathology reports in the test dataset. This algorithm relies on rules defined on data from a single hospital that is specifically dedicated to cancer, but it is based on general processing steps which can be applied to different datasets. Further research will be important to demonstrate the generalizability of the proposed approach on a larger corpus from different hospitals.