Juan Miguel García Gómez
Full Professor at UPV and Head of BDSLab
Marta Durá Hernández
Scientific manager and PhD Student
Carlos Sáez Silvestre
Associate Professor at UPV
Elies Fuster Garcia
Associate Professor at UPV
Sabina Asensio Cuesta
Associate Professor at UPV
Alberto Conejero
Full Professor at UPV
Pablo Ferri Borredá
PhD and Senior Researcher
Vicent Blanes Selva
PhD and Senior Researcher
Ángel Sánchez García
PhD Student
Francisco Javier Gil-Terrón Rodríguez
PhD Student
Carles López Mateu
Researcher
Kevin García Santos
Researcher
María Gómez Mahiques
PhD Student
Víctor Montosa Micó
PhD Student
Daniel Sánchez García
Researcher
María José Cardona Cubells
Administrative Officer
David Fernández Narro
Researcher
Toni Blasco Calafat
PhD Student
Choni Doñate Martínez
PhD and Senior Researcher
Eduard Artur Chelebian Kocharyan
Miquel Oltra
David Lorente
Jose Muñoz
Raquel Faubel
Ricardo Garcia de León
Nekane Romero
Lexin Zhou
Juan Martínez Miranda
Jorge David Mínguez Fons
María del Mar Álvarez Torres
Javier Juan Albarracín
Rebeca Burgos Panadero
Background: Palliative care is referred to a set of programs for patients that suffer life-limiting illnesses. These programs aim to guarantee a minimum level of quality of life (QoL) for the last stage of life. They are currently based on clinical evaluation of risk of one-year mortality.
Objectives: The main objective of this work is to develop and validate machine-learning based models to predict the exitus of a patient within the next year using data gathered at hospital admission.
Methods: Five machine learning techniques were applied in our study to develop machine-learning predictive models: Support Vector Machines, K-neighbors Classifier, Gradient Boosting Classifier, Random Forest and Multilayer Perceptron. All models were trained and evaluated using the retrospective dataset. The evaluation was performed with five metrics computed by a resampling strategy: Accuracy, the area under the ROC curve, Specificity, Sensitivity, and the Balanced Error Rate.
Results: All models for forecasting one-year mortality achieved an AUC ROC from 0.858 to 0.911. Specifically, Gradient Boosting Classifier was the best model, producing an AUC ROC of 0.911 (CI 95%, 0.911 to 0.912), a sensitivity of 0.858 (CI 95%, 0.856 to 0.86) and a specificity of 0.807 (CI 95%, 0.806 to 0808) and a BER of 0.168 (CI 95%, 0.167 to 0.169).
Conclusions: The analysis of common information at hospital admission combined with machine learning techniques produced models with competitive discriminative power. Our models reach the best results reported in state of the art. These results demonstrate that they can be used as an accurate data-driven palliative care criteria inclusion.
Purpose: To determine if preoperative vascular heterogeneity of glioblastoma is predictive of overall survival of patients undergoing standard-of-care treatment by using an unsupervised multiparametric perfusion-based habitat-discovery algorithm.
Materials and Methods: Preoperative magnetic resonance (MR) imaging including dynamic susceptibility-weighted contrast material-enhanced perfusion studies in 50 consecutive patients with glioblastoma were retrieved. Perfusion parameters of glioblastoma were analyzed and used to automatically draw four reproducible habitats that describe the tumor vascular heterogeneity: high-angiogenic and low-angiogenic regions of the enhancing tumor, potentially tumor-infiltrated peripheral edema, and vasogenic edema. Kaplan-Meier and Cox proportional hazard analyses were conducted to assess the prognostic potential of the hemodynamic tissue signature to predict patient survival.
Results: Cox regression analysis yielded a significant correlation between patients' survival and maximum relative cerebral blood volume (rCBVmax) and maximum relative cerebral blood flow (rCBFmax) in high-angiogenic and low-angiogenic habitats (P < .01, false discovery rate-corrected P < .05). Moreover, rCBFmax in the potentially tumor-infiltrated peripheral edema habitat was also significantly correlated (P < .05, false discovery rate-corrected P < .05). Kaplan-Meier analysis demonstrated significant differences between the observed survival of populations divided according to the median of the rCBVmax or rCBFmax at the high-angiogenic and low-angiogenic habitats (log-rank test P < .05, false discovery rate-corrected P < .05), with an average survival increase of 230 days.
Conclusion: Preoperative perfusion heterogeneity contains relevant information about overall survival in patients who undergo standard-of-care treatment. The hemodynamic tissue signature method automatically describes this heterogeneity, providing a set of vascular habitats with high prognostic capabilities.
Justification: Automatic brain tumor classification by MRS has been under development for more than a decade. Nonetheless, to our knowledge, there are no published evaluations of predictive models with unseen cases that are subsequently acquired in different centers. The multicenter eTUMOUR project (2004-2009), which builds upon previous expertise from the INTERPRET project (2000-2002) has allowed such an evaluation to take place.
Materials and methods: A total of 253 pairwise classifiers for glioblastoma, meningioma, metastasis, and low-grade glial diagnosis were inferred based on 211 SV short TE INTERPRET MR spectra obtained at 1.5 T (PRESS or STEAM, 20-32 ms) and automatically pre-processed. Afterwards, the classifiers were tested with 97 spectra, which were subsequently compiled during eTUMOUR.
Results: In our results based on subsequently acquired spectra, accuracies of around 90% were achieved for most of the pairwise discrimination problems. The exception was for the glioblastoma versus metastasis discrimination, which was below 78%. A more clear definition of metastases may be obtained by other approaches, such as MRSI + MRI.
Conclusions: The prediction of the tumor type of in-vivo MRS is possible using classifiers developed from previously acquired data, in different hospitals with different instrumentation under the same acquisition protocols. This methodology may find application for assisting in the diagnosis of new brain tumor cases and for the quality control of multicenter MRS databases.
Objective: To assess the variability in data distributions among data sources and over time through a case study of a large multisite repository as a systematic approach to data quality (DQ).
Materials and methods: Novel probabilistic DQ control methods based on information theory and geometry are applied to the Public Health Mortality Registry of the Region of Valencia, Spain, with 512 143 entries from 2000 to 2012, disaggregated into 24 health departments. The methods provide DQ metrics and exploratory visualizations for (1) assessing the variability among multiple sources and (2) monitoring and exploring changes with time. The methods are suited to big data and multitype, multivariate, and multimodal data.
Results: The repository was partitioned into 2 probabilistically separated temporal subgroups following a change in the Spanish National Death Certificate in 2009. Punctual temporal anomalies were noticed due to a punctual increment in the missing data, along with outlying and clustered health departments due to differences in populations or in practices.
Discussion: Changes in protocols, differences in populations, biased practices, or other systematic DQ problems affected data variability. Even if semantic and integration aspects are addressed in data sharing infrastructures, probabilistic variability may still be present. Solutions include fixing or excluding data and analyzing different sites or time periods separately. A systematic approach to assessing temporal and multisite variability is proposed.
Conclusion: Multisite and temporal variability in data distributions affects DQ, hindering data reuse, and an assessment of such variability should be a part of systematic DQ procedures.
Summary: We present here Blast2GO (B2G), a research tool designed with the main purpose of enabling Gene Ontology (GO) based data mining on sequence data for which no GO annotation is yet available. B2G joints in one application GO annotation based on similarity searches with statistical analysis and highlighted visualization on directed acyclic graphs. This tool offers a suitable platform for functional genomics research in non-model species. B2G is an intuitive and interactive desktop application that allows monitoring and comprehension of the whole annotation and analysis process.
Abstract: Actigraphy devices have been successfully used as effective tools in the treatment of diseases such as sleep disorders or major depression. Although several efforts have been made in recent years to develop smaller and more portable devices, the features necessary for the continuous monitoring of outpatients require a less intrusive, obstructive and stigmatizing acquisition system. A useful strategy to overcome these limitations is based on adapting the monitoring system to the patient lifestyle and behavior by providing sets of different sensors that can be worn simultaneously or alternatively. This strategy offers to the patient the option of using one device or other according to his/her particular preferences. However this strategy requires a robust multi-sensor fusion methodology capable of taking maximum profit from all of the recorded information. With this aim, this study proposes two actigraphy fusion models including centralized and distributed architectures based on artificial neural networks. These novel fusion methods were tested both on synthetic datasets and real datasets, providing a parametric characterization of the models' behavior, and yielding results based on real case applications. The results obtained using both proposed fusion models exhibit good performance in terms of robustness to signal degradation, as well as a good behavior in terms of the dependence of signal quality on the number of signals fused. The distributed and centralized fusion methods reduce the mean averaged error of the original signals to 44% and 46% respectively when using simulated datasets. The proposed methods may therefore facilitate a less intrusive and more dependable way of acquiring valuable monitoring information from outpatients.
Objective: The main goal of this paper is to obtain a classification model based on feed-forward multilayer perceptrons in order to improve postpartum depression prediction during the 32 weeks after childbirth with a high sensitivity and specificity and to develop a tool to be integrated in a decision support system for clinicians.
Materials and methods: Multilayer perceptrons were trained on data from 1397 women who had just given birth, from seven Spanish general hospitals, including clinical, environmental and genetic variables. A prospective cohort study was made just after delivery, at 8 weeks and at 32 weeks after delivery. The models were evaluated with the geometric mean of accuracies using a hold-out strategy.
Results: Multilayer perceptrons showed good performance (high sensitivity and specificity) as predictive models for postpartum depression.
Conclusions: The use of these models in a decision support system can be clinically evaluated in future work. The analysis of the models by pruning leads to a qualitative interpretation of the influence of each variable in the interest of clinical protocols.
We investigate and apply novel methods to measure, describe and control data quality and variability for trustworthy use of biomedical data. Our 9-dimensional DQ framework and specialized methods for assessing temporal and multi-source variability complement database, machine learning and visual analytics approaches. Being robust to data quality and dataset shifts, we aim for reliable and explainable artificial intelligence for real world data.
Artificial Intelligence is revolutionizing medical imaging, becoming an indisputable tool for modern medicine. We investigate innovative Deep Learning solutions to extract valuable knowledge from images to help addressing complex clinical problems. Our habitats-based imaging technology has demonstrated strong correlations with relevant clinical outcomes in patients with glioblastoma, unlocking new possibilities in early therapy planning support.
Our research line in decision support is focused on the development of predictive models and Clinical Decision Support Systems (CDSS). From rule-based systems implementing international diabetes guidelines to deep learning for emergency medical call incidents classification and machine learning models to assess palliative care needs. Our goal is to provide decision support tools to physicians and health experts concerning several health issues.
We investigate and apply novel methods to measure, describe and control data quality and variability for trustworthy use of biomedical data. Our 9-dimensional DQ framework and specialized methods for assessing temporal and multi-source variability complement database, machine learning and visual analytics approaches. Being robust to data quality and dataset shifts, we aim for reliable and explainable artificial intelligence for real world data.
Artificial Intelligence is revolutionizing medical imaging, becoming an indisputable tool for modern medicine. We investigate innovative Deep Learning solutions to extract valuable knowledge from images to help addressing complex clinical problems. Our habitats-based imaging technology has demonstrated strong correlations with relevant clinical outcomes in patients with glioblastoma, unlocking new possibilities in early therapy planning support.
Our research line in decision support is focused on the development of predictive models and Clinical Decision Support Systems (CDSS). From rule-based systems implementing international diabetes guidelines to deep learning for emergency medical call incidents classification and machine learning models to assess palliative care needs. Our goal is to provide decision support tools to physicians and health experts concerning several health issues.
Error loading projects.... Please try to reload the web page.
Error loading individual grants.... Please try to reload the web page.
Error loading publications.... Please try to reload the web page.
ONCOhabitats
ONCOhabitats is an on-line open access platform for glioblastoma heterogeneity assessment by means of MRI analysis through machine learning techniques [1]. ONCOhabitats provides two services for untreated glioblastomas: 1) malignant tissue segmentation, and 2) vascular heterogeneity assessment through the HTS method. The malignant tissue segmentation employs state-of-the-art Deep Learning techniques to delineate high grade glioma tissues such as the enhancing tumor, surrounding edema or the necrotic core, with a performance score of 90% of confidence with respect to professional manual segmentations provided by expert radiologist [1]. Vascular heterogeneity assessment of the tumor is conducted through the Hemodynamic Tissue Signature [2], which analyzes perfusion MIR markers to automatically draw four reproducible habitats that describe the tumor vascular heterogeneity: the High Angiogenic Tumor (HAT) and Low Angiogenic Tumor (LAT) habitats of the enhancing tumor, the potentially tumor Infiltrated Peripheral Edema (IPE) and the Vasogenic Peripheral Edema (VPE). These habitats have demonstrated early prognostic capabilities and strong correlations with patient overall survival in large international multi-center studies [ 3]. ONCOhabitats platform also provides automated radiological reports for each service, including volumetries and perfusion measurements of the different regions of the lesion. Finally, ONCOhabitats gives access to the scientific community to a computational cluster capable to process about 300 cases per day.
Wakamola
Wakamola is a chatbot on the Telegram Platform developed in the context of the CrowdHealth project, in the use case of obesity identification. The main goal of Wakamola is to ask questions to the user about his/her habits, nutrition patterns and exercise activities. This function allows us to collect information to help our research, and at the same time, makes the user think about his/her conducts in an "auto evaluation" way. Also, Wakamola provides a score (Wakascore) which is an approximate representation of the user's health status based on the responses received. At this moment, we are working to make Wakamola available on more platforms.
Crowdhealth
BDSLab's HAT (health analytics tool) for the CrowdHealth project is a series of models to identify obesity on the general population using clinical data that does not contain anthropometric measures. The models are based on the PU Learning paradigm since the main issue with this task is the lack of obesity identification among the EHR (Electronic Health Records). This tool is wrapped within a REST API which uses the Policy Development Toolkit from CrowdHealth project to launch the tool and visualise its results
Contact: viblasel@upv.es
Lalaby
The lalaby smartphone app allows the monitoring of patients through the motion, location, conversation detection and data usage sensors of smartphones.
Lalaby also allows the incorporation of questionnaires such as the EORTC QLQ-C30 (European Organization for Research and Treatment of Cancer- QLQ-C30) which is a validated questionnaire widely used in the clinical field for the evaluation of the quality of life in cancer patients, as well as other questionnaires related to chronic diseases with which the lalaby sensor records are to be correlated and considered of interest in each case study, such as the study of quality of life in patients with migraine, depression, Alzheimer's disease, etc.
The lalaby app allows obtaining patterns of behavior in order to relate them to indicators of the patient's quality of life. In this way, lalaby allows the evaluation of the daily quality of life of cancer patients under treatment from the information collected automatically with their mobile phones.
Contact: sasensio@upv.es
COVID-19 SDE Tool
The COVID-19 infectious disease has led since December 2019 to a worldwide pandemic which is still under control measures. Researchers worldwide are making huge efforts aiming to a comprehensive understanding of the COVID-19 and related healthcare treatments. This work shows the preliminary results of a Machine Learning (ML) approach to identify subgroups of COVID-19 patients based on their symptoms and comorbidities, aiming to a better understanding of variability of severity patterns. In this work, we particularly address the variability (or heterogeneity) between distinct sources populating the research repositories, given the potential impact that this variability may have in data science and the generalization of its results.
EHRTemporalVariability
Variability in healthcare processes, protocols or due to the system or human biases, can be a potential bias for the reuse of Electronic Health Records (EHRs), where unexpected batch-effects can be introduced. EHRtemporalVariability is an open-source R-package and Shiny-app for exploring and uncovering the effects of time in the statistical distributions of EHRs, namely dataset shifts. EHRtemporalVariability batches and visualizes EHRs temporal-evolution through dynamic heatmaps and non-parametric-information-geometry plots for coded, numerical and multivariate EHR data. Explore a case study of the app capabilities with the US National Hospital Discharge Survey open dataset!
112
112 demo consists on a graphical user interface linked to a predictive model. It emulates the data collection process during a medical emergency call and classifies the reported incident by means of three different but complementary labels: life-threatening event (yes/no), admissible response delay (undelayable, minutes, hours, days) and jurisdiction (emergency system/primary care). The underlying predictive model is composed by a deep multitask modular neural network, website
Deep multitask ensemble classification of emergency medical call incidents combining multimodal data
ALCOA+
The ALCOA+ web application is based on the SPuMoNI project, related to the task of easing auditing production reports in the manufacturing pharmaceutical industry through the use of automatic analysis using the ALCOA+ data integrity principles.
This web application analyzes the data quality from a data structure in order to determine the scores for every ALCOA+ principle.
There are two categories:
Single element data analysis: it is computed the ALCOA+ scores within the range of 0 to 100 and shown through a plot. The overall score is also plotted. Data can be visualized through a visualizer webpage. This also is useful for the deep analysis, where it is shown to the user every data part that has something wrong.
Multiple elements data analysis: when it is analyzed the behaviour of multiple elements it is plotted a heatmap showing the temporal variability of every principle and the overall measure. For every principle is plotted a Statistical Process Control (SPC) Chart, where it is visualized the scores and the limits. Clicking on a dot on the SPC charts allows analyzing the element individually.
Palliative Care Assessment Tool
The BDSLab's demo for the InAdvance problem consists of three predictive models (one-year mortality classification, mortality regression and six-months frailty index regression) to assess the palliative care needs among the hospital admitted patients. The models have been integrated into a website that allows their use and evaluation from single evaluations introducing a patient's data through a form to more complex operations.
AgrofoodAI
AgrofoodAI is a tool based on a prediction model of the humidity of the ground of a specific field for field irrigation planning.
The prediction model is a deep neural network (U-Net) which takes as input multiple time data series and outputs the humidity of the ground for the next week.
This model takes as input: time data series from the last week like ground humidity, ground temperature, ambient humidity, irrigation planning and rain, and other time series from the current week such as rain prediction or the irrigation planning that is going to be used.
This tool allows the user to predict the behaviour of the field's humidity in order to modify or not the irrigation planning to be used.
COVID Calculator
This is a prototype tool that calculates the 30-day mortality risk after emergency admission for suspected or confirmed SARS-CoV-2 infection in patients over 50 years.
Developed in the context of the project 'Severity Subgroup Discovery and Classification on COVID-19 Real World Data through Machine Learning and Data Quality assessment (SUBCOVERWD-19)' funded by Fondo Supera COVID-19 by CRUE (Conferencia de Rectores de las Universidades Españolas) - Santander Universidades (Santander Bank), from Oct 2020 to Dec 2021. In collaboration with Instituto de Investigación IMAS12 - Hospital 12 de Octubre and Instituto de Investigación Sanitaria INCLIVA - Hospital Clínico Universitario de Valencia
Send us a message through the website or the following links:
Ilustraciones realizadas dentro del proyecto "Arte para la ciencia", con el apoyo de:
