RESUMEN
In recent years, the application of artificial intelligence (AI) in the automotive industry has led to the development of intelligent systems focused on road safety, aiming to improve protection for drivers and pedestrians worldwide to reduce the number of accidents yearly. One of the most critical functions of these systems is pedestrian detection, as it is crucial for the safety of everyone involved in road traffic. However, pedestrian detection goes beyond the front of the vehicle; it is also essential to consider the vehicle's rear since pedestrian collisions occur when the car is in reverse drive. To contribute to the solution of this problem, this research proposes a model based on convolutional neural networks (CNN) using a proposed one-dimensional architecture and the Inception V3 architecture to fuse the information from the backup camera and the distance measured by the ultrasonic sensors, to detect pedestrians when the vehicle is reversing. In addition, specific data collection was performed to build a database for the research. The proposed model showed outstanding results with 99.85% accuracy and 99.86% correct classification performance, demonstrating that it is possible to achieve the goal of pedestrian detection using CNN by fusing two types of data.
RESUMEN
Driver identification refers to the process whose primary purpose is identifying the person behind the steering wheel using collected information about the driver him/herself. The constant monitoring of drivers through sensors generates great benefits in advanced driver assistance systems (ADAS), to learn more about the behavior of road users. Currently, there are many research works that address the subject in search of creating intelligent models that help to identify vehicle users in an efficient and objective way. However, the different methodologies proposed to create these models are based on data generated from sensors that include different vehicle brands on routes established in real environments, which, although they provide very important information for different purposes, in the case of driver identification, there may be a certain degree of bias due to the different situations in which the route environment may change. The proposed method seeks to intelligently and objectively select the most outstanding statistical features from motor activity generated in the main elements of the vehicle with genetic algorithms for driver identification, this process being newer than those established by the state-of-the-art. The results obtained from the proposal were an accuracy of 90.74% to identify two drivers and 62% for four, using a Random Forest Classifier (RFC). With this, it can be concluded that a comprehensive selection of features can greatly optimize the identification of drivers.
Asunto(s)
Conducción de Automóvil , Humanos , Masculino , Accidentes de Tránsito , Bosques Aleatorios , Aprendizaje , Actividad MotoraRESUMEN
Automatic identification of human facial expressions has many potential applications in today's connected world, from mental health monitoring to feedback for onscreen content or shop windows and sign-language prosodic identification. In this work we use visual information as input, namely, a dataset of face points delivered by a Kinect device. The most recent work on facial expression recognition uses Machine Learning techniques, to use a modular data-driven path of development instead of using human-invented ad hoc rules. In this paper, we present a Machine-Learning based method for automatic facial expression recognition that leverages information fusion architecture techniques from our previous work and soft voting. Our approach shows an average prediction performance clearly above the best state-of-the-art results for the dataset considered. These results provide further evidence of the usefulness of information fusion architectures rather than adopting the default ML approach of features aggregation.
Asunto(s)
Reconocimiento Facial , Cara , Expresión Facial , Humanos , Aprendizaje Automático , PolíticaRESUMEN
Background: The coronavirus disease (COVID-19) is an infectious disease caused by the SARS-CoV-2 virus and is responsible for nearly 6 million deaths worldwide in the past 2 years. Machine learning (ML) models could help physicians in identifying high-risk individuals. Objectives: To study the use of ML models for COVID-19 prediction outcomes using clinical data and a combination of clinical and metabolic data, measured in a metabolomics facility from a public university. Methods: A total of 154 patients were included in the study. "Basic profile" was considered with clinical and demographic variables (33 variables), whereas in the "extended profile," metabolomic and immunological variables were also considered (156 characteristics). A selection of features was carried out for each of the profiles with a genetic algorithm (GA) and random forest models were trained and tested to predict each of the stages of COVID-19. Results: The model based on extended profile was more useful in early stages of the disease. Models based on clinical data were preferred for predicting severe and critical illness and death. ML detected trimethylamine N-oxide, lipid mediators, and neutrophil/lymphocyte ratio as important variables. Conclusions: ML and GAs provided adequate models to predict COVID-19 outcomes in patients with different severity grades.
Asunto(s)
COVID-19 , SARS-CoV-2 , Humanos , COVID-19/diagnóstico , Algoritmos , Pronóstico , Aprendizaje AutomáticoRESUMEN
Worldwide, motor vehicle accidents are one of the leading causes of death, with alcohol-related accidents playing a significant role, particularly in child death. Aiming to aid in the prevention of this type of accidents, a novel non-invasive method capable of detecting the presence of alcohol inside a motor vehicle is presented. The proposed methodology uses a series of low-cost alcohol MQ3 sensors located inside the vehicle, whose signals are stored, standardized, time-adjusted, and transformed into 5 s window samples. Statistical features are extracted from each sample and a feature selection strategy is carried out using a genetic algorithm, and a forward selection and backwards elimination methodology. The four features derived from this process were used to construct an SVM classification model that detects presence of alcohol. The experiments yielded 7200 samples, 80% of which were used to train the model. The rest were used to evaluate the performance of the model, which obtained an area under the ROC curve of 0.98 and a sensitivity of 0.979. These results suggest that the proposed methodology can be used to detect the presence of alcohol and enforce prevention actions.
Asunto(s)
Conducción de Automóvil , Conducir bajo la Influencia , Accidentes de Tránsito/prevención & control , Algoritmos , Niño , Humanos , Vehículos a MotorRESUMEN
Among the current challenges of the Smart City, traffic management and maintenance are of utmost importance. Road surface monitoring is currently performed by humans, but the road surface condition is one of the main indicators of road quality, and it may drastically affect fuel consumption and the safety of both drivers and pedestrians. Abnormalities in the road, such as manholes and potholes, can cause accidents when not identified by the drivers. Furthermore, human-induced abnormalities, such as speed bumps, could also cause accidents. In addition, while said obstacles ought to be signalized according to specific road regulation, they are not always correctly labeled. Therefore, we developed a novel method for the detection of road abnormalities (i.e., speed bumps). This method makes use of a gyro, an accelerometer, and a GPS sensor mounted in a car. After having the vehicle cruise through several streets, data is retrieved from the sensors. Then, using a cross-validation strategy, a genetic algorithm is used to find a logistic model that accurately detects road abnormalities. The proposed model had an accuracy of 0.9714 in a blind evaluation, with a false positive rate smaller than 0.018, and an area under the receiver operating characteristic curve of 0.9784. This methodology has the potential to detect speed bumps in quasi real-time conditions, and can be used to construct a real-time surface monitoring system.
RESUMEN
Human Activity Recognition (HAR) is one of the main subjects of study in the areas of computer vision and machine learning due to the great benefits that can be achieved. Examples of the study areas are: health prevention, security and surveillance, automotive research, and many others. The proposed approaches are carried out using machine learning techniques and present good results. However, it is difficult to observe how the descriptors of human activities are grouped. In order to obtain a better understanding of the the behavior of descriptors, it is important to improve the abilities to recognize the human activities. This paper proposes a novel approach for the HAR based on acoustic data and similarity networks. In this approach, we were able to characterize the sound of the activities and identify those activities looking for similarity in the sound pattern. We evaluated the similarity of the sounds considering mainly two features: the sound location and the materials that were used. As a result, the materials are a good reference classifying the human activities compared with the location.
Asunto(s)
Actividades Humanas , Acústica , Humanos , Reconocimiento de Normas Patrones Automatizadas , Procesamiento de Señales Asistido por Computador , SonidoRESUMEN
In this paper, we present the development of an infrastructure-less indoor location system (ILS), which relies on the use of a microphone, a magnetometer and a light sensor of a smartphone, all three of which are essentially passive sensors, relying on signals available practically in any building in the world, no matter how developed the region is. In our work, we merge the information from those sensors to estimate the user's location in an indoor environment. A multivariate model is applied to find the user's location, and we evaluate the quality of the resulting model in terms of sensitivity and specificity. Our experiments were carried out in an office environment during summer and winter, to take into account changes in light patterns, as well as changes in the Earth's magnetic field irregularities. The experimental results clearly show the benefits of using the information fusion of multiple sensors when contrasted with the use of a single source of information.
RESUMEN
User indoor positioning has been under constant improvement especially with the availability of new sensors integrated into the modern mobile devices, which allows us to exploit not only infrastructures made for everyday use, such as WiFi, but also natural infrastructure, as is the case of natural magnetic field. In this paper we present an extension and improvement of our current indoor localization model based on the feature extraction of 46 magnetic field signal features. The extension adds a feature selection phase to our methodology, which is performed through Genetic Algorithm (GA) with the aim of optimizing the fitness of our current model. In addition, we present an evaluation of the final model in two different scenarios: home and office building. The results indicate that performing a feature selection process allows us to reduce the number of signal features of the model from 46 to 5 regardless the scenario and room location distribution. Further, we verified that reducing the number of features increases the probability of our estimator correctly detecting the user's location (sensitivity) and its capacity to detect false positives (specificity) in both scenarios.
RESUMEN
Type 2 diabetes mellitus (T2DM) is one of the most common metabolic diseases in the world and poses a significant public health challenge. Early detection and management of this metabolic disorder is crucial to prevent complications and improve outcomes. This paper aims to find core differences in male and female markers to detect T2DM by their clinic and anthropometric features, seeking out ranges in potential biomarkers identified to provide useful information as a pre-diagnostic tool whie excluding glucose-related biomarkers using machine learning (ML) models. We used a dataset containing clinical and anthropometric variables from patients diagnosed with T2DM and patients without TD2M as control. We applied feature selection with three different techniques to identify relevant biomarker models: an improved recursive feature elimination (RFE) evaluating each set from all the features to one feature with the Akaike information criterion (AIC) to find optimal outputs; Least Absolute Shrinkage and Selection Operator (LASSO) with glmnet; and Genetic Algorithms (GA) with GALGO and forward selection (FS) applied to GALGO output. We then used these for comparison with the AIC to measure the performance of each technique and collect the optimal set of global features. Then, an implementation and comparison of five different ML models was carried out to identify the most accurate and interpretable one, considering the following models: logistic regression (LR), artificial neural network (ANN), support vector machine (SVM), k-nearest neighbors (KNN), and nearest centroid (Nearcent). The models were then combined in an ensemble to provide a more robust approximation. The results showed that potential biomarkers such as systolic blood pressure (SBP) and triglycerides are together significantly associated with T2DM. This approach also identified triglycerides, cholesterol, and diastolic blood pressure as biomarkers with differences between male and female actors that have not been previously reported in the literature. The most accurate ML model was selection with RFE and random forest (RF) as the estimator improved with the AIC, which achieved an accuracy of 0.8820. In conclusion, this study demonstrates the potential of ML models in identifying potential biomarkers for early detection of T2DM, excluding glucose-related biomarkers as well as differences between male and female anthropometric and clinic profiles. These findings may help to improve early detection and management of the T2DM by accounting for differences between male and female subjects in terms of anthropometric and clinic profiles, potentially reducing healthcare costs and improving personalized patient attention. Further research is needed to validate these potential biomarkers ranges in other populations and clinical settings.
RESUMEN
The occurrence of fungi is cosmopolitan, and while some mushroom species are beneficial to human health, others can be toxic and cause illness problems. This study aimed to analyze the organoleptic, ecological, and morphological characteristics of a group of fungal specimens and identify the most significant features to develop models for fungal toxicity classification using genetic algorithms and LASSO regression. The results of the study indicated that odor, spore print color, and habitat were the most significant characteristics identified by the genetic algorithm GALGO. Meanwhile, odor, gill size, stalk shape, and twelve other features were the relevant characteristics identified by LASSO regression. The importance score of the odor variable was 99.99%, gill size obtained 73.7%, stalk shape scored 39.9%, and the remaining variables did not score higher than 18%. Logistic regression, k-nearest neighbor (KNN), and XG-Boost classification algorithms were used to develop models using the features selected by both GALGO and LASSO. The models were evaluated using sensitivity, specificity, and accuracy metrics. The models with the highest AUC values were XGBoost, with a maximum value of 0.99 using the features selected by LASSO, followed by KNN with a maximum value of 0.99. The GALGO selection resulted in a maximum AUC of 0.98 in KNN and XGBoost. The models developed in this study have the potential to aid in the accurate identification of toxic fungi, which can prevent health problems caused by their consumption.
Asunto(s)
Agaricus , Humanos , Agaricus/genética , Algoritmos , Benchmarking , Análisis por Conglomerados , Aprendizaje AutomáticoRESUMEN
The development of medical diagnostic models to support healthcare professionals has witnessed remarkable growth in recent years. Among the prevalent health conditions affecting the global population, diabetes stands out as a significant concern. In the domain of diabetes diagnosis, machine learning algorithms have been widely explored for generating disease detection models, leveraging diverse datasets primarily derived from clinical studies. The performance of these models heavily relies on the selection of the classifier algorithm and the quality of the dataset. Therefore, optimizing the input data by selecting relevant features becomes essential for accurate classification. This research presents a comprehensive investigation into diabetes detection models by integrating two feature selection techniques: the Akaike information criterion and genetic algorithms. These techniques are combined with six prominent classifier algorithms, including support vector machine, random forest, k-nearest neighbor, gradient boosting, extra trees, and naive Bayes. By leveraging clinical and paraclinical features, the generated models are evaluated and compared to existing approaches. The results demonstrate superior performance, surpassing accuracies of 94%. Furthermore, the use of feature selection techniques allows for working with a reduced dataset. The significance of feature selection is underscored in this study, showcasing its pivotal role in enhancing the performance of diabetes detection models. By judiciously selecting relevant features, this approach contributes to the advancement of medical diagnostic capabilities and empowers healthcare professionals in making informed decisions regarding diabetes diagnosis and treatment.
Asunto(s)
Algoritmos , Diabetes Mellitus , Humanos , Teorema de Bayes , Aprendizaje Automático , Diabetes Mellitus/diagnóstico , Bosques AleatoriosRESUMEN
The gastrointestinal (GI) tract can be affected by different diseases or lesions such as esophagitis, ulcers, hemorrhoids, and polyps, among others. Some of them can be precursors of cancer such as polyps. Endoscopy is the standard procedure for the detection of these lesions. The main drawback of this procedure is that the diagnosis depends on the expertise of the doctor. This means that some important findings may be missed. In recent years, this problem has been addressed by deep learning (DL) techniques. Endoscopic studies use digital images. The most widely used DL technique for image processing is the convolutional neural network (CNN) due to its high accuracy for modeling complex phenomena. There are different CNNs that are characterized by their architecture. In this article, four architectures are compared: AlexNet, DenseNet-201, Inception-v3, and ResNet-101. To determine which architecture best classifies GI tract lesions, a set of metrics; accuracy, precision, sensitivity, specificity, F1-score, and area under the curve (AUC) were used. These architectures were trained and tested on the HyperKvasir dataset. From this dataset, a total of 6,792 images corresponding to 10 findings were used. A transfer learning approach and a data augmentation technique were applied. The best performing architecture was DenseNet-201, whose results were: 97.11% of accuracy, 96.3% sensitivity, 99.67% specificity, and 95% AUC.
Asunto(s)
Aprendizaje Profundo , Redes Neurales de la Computación , Tracto Gastrointestinal/diagnóstico por imagen , Endoscopía Gastrointestinal , Diagnóstico por Computador/métodosRESUMEN
The escalating prevalence of Type 2 Diabetes (T2D) represents a substantial burden on global healthcare systems, especially in regions such as Mexico. Existing diagnostic techniques, although effective, often require invasive procedures and labor-intensive efforts. The promise of artificial intelligence and data science for streamlining and enhancing T2D diagnosis is well-recognized; however, these advancements are frequently constrained by the limited availability of comprehensive patient datasets. To mitigate this challenge, the present study investigated the efficacy of Generative Adversarial Networks (GANs) for augmenting existing T2D patient data, with a focus on a Mexican cohort. The researchers utilized a dataset of 1019 Mexican nationals, divided into 499 non-diabetic controls and 520 diabetic cases. GANs were applied to create synthetic patient profiles, which were subsequently used to train a Random Forest (RF) classification model. The study's findings revealed a notable improvement in the model's diagnostic accuracy, validating the utility of GAN-based data augmentation in a clinical context. The results bear significant implications for enhancing the robustness and reliability of Machine Learning tools in T2D diagnosis and management, offering a pathway toward more timely and effective patient care.
RESUMEN
The aim of this study was to analyze molecules associated with regulatory immune response in unvaccinated, recovered COVID-19 patients with and without diabetes mellitus (DM) and hypertension (HTN). We determined anti-SARS-CoV-2 nucleocapsid IgG in plasma by electrochemiluminescence immunoassay. The levels of sCD40, TGF-ß, IL-10, and sCTLA-4 were assessed by ELISA in the serum of the subjects, as well as in healthy donors. We observed that only half of the subjects in the non-comorbid group produced antibodies, whereas all subjects in comorbid groups were IgG-positive for the anti-SARS-CoV-2 nucleocapsid. High levels of sCTL-4 were observed in the non-comorbid group, and the level of IL-10 was observed to increase in seropositive subjects without comorbidities. TGF-ß concentration was similar in all groups studied. Finally, sCD40 decreased in the comorbid group. In conclusion, our results suggest that comorbidities such as DM and HTN alter the production of co-stimulatory inhibitory molecules sCTLA-4 and sCD40 in subjects recovering from mild COVID-19. The alterations observed here were independent of seropositivity, suggesting an effective humoral immune response against COVID-19 separate from the levels of co-stimulatory inhibitory molecules.
RESUMEN
Depression is a common illness worldwide, affecting an estimated 3.8% of the population, including 5% of all adults, in particular, 5.7% of adults over 60 years of age. Unfortunately, at present, the ways to evaluate different mental disorders, like the Montgomery-Åsberg depression rating scale (MADRS) and observations, need a great effort, on part of specialists due to the lack of availability of patients to obtain the necessary information to know their conditions and to detect illness such as depression in an objective way. Based on data analysis and artificial intelligence techniques, like Convolutional Neural Network (CNN), it is possible to classify a person, from the mental status examination, into two classes. Moreover, it is beneficial to observe how the data of these two classes are similar in different time intervals. In this study, a motor activity database was used, from which the readings of 55 subjects of study (32 healthy and 23 with some degree of depression) were recorded with a small wrist-worn accelerometer to detect the peak amplitude of movement acceleration and generate a transient voltage signal proportional to the rate of acceleration. Motor activity data were selected per patient in time-lapses of one day for seven days (one week) in one-minute intervals. The data were pre-processed to be given to a two-dimensional convolutional network (2D-CNN), where each record of motor activity per minute was represented as a pixel of an image. The proposed model is capable of detecting depression in real-time (if this is implemented in a mobile device such as a smartwatch) with low computational cost and accuracy of 76.72% In summary, the model shows promising abilities to detect possible cases of depression, providing a helpful resource to identify the condition and be able to take the appropriate follow-up for the patient.
RESUMEN
Breast cancer is the most common cancer among women worldwide, after lung cancer. However, early detection of breast cancer can help to reduce death rates in breast cancer patients and also prevent cancer from spreading to other parts of the body. This work proposes a new method to design a bio-marker integrating Bayesian predictive models, pyRadiomics System and genetic algorithms to classify the benign and malignant lesions. The method allows one to evaluate two types of images: The radiologist-segmented lesion, and a novel automated breast cancer detection by the analysis of the whole breast. The results demonstrate only a difference of 12% of effectiveness for the cases of calcification between the radiologist generated segmentation and the automatic whole breast analysis, and a 25% of difference between the lesion and the breast for the cases of masses. In addition, our approach was compared against other proposed methods in the literature, providing an AUC = 0.86 for the analysis of images with lesions in breast calcification, and AUC = 0.96 for masses.
RESUMEN
Major depressive disorder (MDD) is the most recurrent mental illness globally, affecting approximately 5% of adults. Furthermore, according to the National Institute of Mental Health (NIMH) of the U.S., calculating an actual schizophrenia prevalence rate is challenging because of this illness's underdiagnosis. Still, most current global metrics hover between 0.33% and 0.75%. Machine-learning scientists use data from diverse sources to analyze, classify, or predict to improve the psychiatric attention, diagnosis, and treatment of MDD, schizophrenia, and other psychiatric conditions. Motor activity data are gaining popularity in mental illness diagnosis assistance because they are a cost-effective and noninvasive method. In the knowledge discovery in databases (KDD) framework, a model to classify depressive and schizophrenic patients from healthy controls is constructed using accelerometer data. Taking advantage of the multiple sleep disorders caused by mental disorders, the main objective is to increase the model's accuracy by employing only data from night-time activity. To compare the classification between the stages of the day and improve the accuracy of the classification, the total activity signal was cut into hourly time lapses and then grouped into subdatasets depending on the phases of the day: morning (06:00-11:59), afternoon (12:00-17:59), evening (18:00-23:59), and night (00:00-05:59). Random forest classifier (RFC) is the algorithm proposed for multiclass classification, and it uses accuracy, recall, precision, the Matthews correlation coefficient, and F1 score to measure its efficiency. The best model was night-featured data and RFC, with 98% accuracy for the classification of three classes. The effectiveness of this experiment leads to less monitoring time for patients, reducing stress and anxiety, producing more efficient models, using wearables, and increasing the amount of data.
RESUMEN
Sudden infant death syndrome (SIDS) represents the leading cause of death in under one year of age in developing countries. Even in our century, its etiology is not clear, and there is no biomarker that is discriminative enough to predict the risk of suffering from it. Therefore, in this work, taking a public dataset on the lipidomic profile of babies who died from this syndrome compared to a control group, a univariate analysis was performed using the Mann-Whitney U test, with the aim of identifying the characteristics that enable discriminating between both groups. Those characteristics with a p-value less than or equal to 0.05 were taken; once these characteristics were obtained, classification models were implemented (random forests (RF), logistic regression (LR), support vector machine (SVM) and naive Bayes (NB)). We used seventy percent of the data for model training, subjecting it to a cross-validation (k = 5) and later submitting to validation in a blind test with 30% of the remaining data, which allows simulating the scenario in real life-that is, with an unknown population for the model. The model with the best performance was RF, since in the blind test, it obtained an AUC of 0.9, specificity of 1, and sensitivity of 0.8. The proposed model provides the basis for the construction of a SIDS risk prediction computer tool, which will contribute to prevention, and proposes lines of research to deal with this pathology.
RESUMEN
Type 2 diabetes mellitus (T2DM) represents one of the biggest health problems in Mexico, and it is extremely important to early detect this disease and its complications. For a noninvasive detection of T2DM, a machine learning (ML) approach that uses ensemble classification models with dichotomous output that is also fast and effective for early detection and prediction of T2D can be used. In this article, an ensemble technique by hard voting is designed and implemented using generalized linear regression (GLM), support vector machines (SVM) and artificial neural networks (ANN) for the classification of T2DM patients. In the materials and methods as a first step, the data is balanced, standardized, imputed and integrated into the three models to classify the patients in a dichotomous result. For the selection of features, an implementation of LASSO is developed, with a 10-fold cross-validation and for the final validation, the Area Under the Curve (AUC) is used. The results in LASSO showed 12 features, which are used in the implemented models to obtain the best possible scenario in the developed ensemble model. The algorithm with the best performance of the three is SVM, this model obtained an AUC of 92% ± 3%. The ensemble model built with GLM, SVM and ANN obtained an AUC of 90% ± 3%.