Risk of Bias in Chest Radiography Deep Learning Foundation Models.

Purpose: To analyze a recently published chest radiography foundation model for the presence of biases that could lead to subgroup performance disparities across biologic sex and race. Materials and Methods: This Health Insurance Portability and Accountability Act-compliant retrospective study used 127 118 chest radiographs from 42 884 patients (mean age, 63 years ± 17 [SD]; 23 623 male, 19 261 female) from the CheXpert dataset that were collected between October 2002 and July 2017. To determine the presence of bias in features generated by a chest radiography foundation model and baseline deep learning model, dimensionality reduction methods together with two-sample Kolmogorov-Smirnov tests were used to detect distribution shifts across sex and race. A comprehensive disease detection performance analysis was then performed to associate any biases in the features to specific disparities in classification performance across patient subgroups. Results: Ten of 12 pairwise comparisons across biologic sex and race showed statistically significant differences in the studied foundation model, compared with four significant tests in the baseline model. Significant differences were found between male and female (P < .001) and Asian and Black (P < .001) patients in the feature projections that primarily capture disease. Compared with average model performance across all subgroups, classification performance on the "no finding" label decreased between 6.8% and 7.8% for female patients, and performance in detecting "pleural effusion" decreased between 10.7% and 11.6% for Black patients. Conclusion: The studied chest radiography foundation model demonstrated racial and sex-related bias, which led to disparate performance across patient subgroups; thus, this model may be unsafe for clinical applications.Keywords: Conventional Radiography, Computer Application-Detection/Diagnosis, Chest Radiography, Bias, Foundation Models Supplemental material is available for this article. Published under a CC BY 4.0 license.See also commentary by Czum and Parr in this issue.

Evaluation of a 3-Item Health Index in Predicting Mortality Risk: A 12-Year Follow-Up Study.

This study was carried out using a large cohort (N = 4265; 416 deceased) of older, community-dwelling adults from The Irish Longitudinal Study on Ageing (TILDA). The study compared the performance of a new 3-item health index (HI) with two existing measures, the 32-item frailty index (FI) and the frailty phenotype (FP), in predicting mortality risk. The HI was based on the objective measurement of resting-state systolic blood pressure sample entropy, sustained attention reaction time performance, and usual gait speed. Mortality data from a 12-year follow up period were analyzed using Cox proportional regression. All data processing was performed using MATLAB and statistical analysis using STATA 15.1. The HI showed good discriminatory power (AUC = 0.68) for all-cause mortality, similar to FI (AUC = 0.68) and superior to FP (AUC = 0.60). The HI classified participants into Low-Risk (84%), Medium-Risk (15%), and High-Risk (1%) groups, with the High-Risk group showing a significant hazard ratio (HR) of 5.91 in the unadjusted model and 2.06 in the fully adjusted model. The HI also exhibited superior predictive performance for cardiovascular and respiratory deaths (AUC = 0.74), compared with FI (AUC = 0.70) and FP (AUC = 0.64). The HI High-Risk group had the highest HR (15.10 in the unadjusted and 5.61 in the fully adjusted models) for cardiovascular and respiratory mortality. The HI remained a significant predictor of mortality even after comprehensively adjusting for confounding variables. These findings demonstrate the effectiveness of the 3-item HI in predicting 12-year mortality risk across different causes of death. The HI performed similarly to FI and FP for all-cause mortality but outperformed them in predicting cardiovascular and respiratory deaths. Its ability to classify individuals into risk groups offers a practical approach for clinicians and researchers. Additionally, the development of a user-friendly MATLAB App facilitates its implementation in clinical settings. Subject to external validation in clinical research settings, the HI can be more useful than existing frailty measures in the prediction of cardio-respiratory risk.

FoCo: A Shiny App for Formative Assessment using Cognitive Diagnosis Modeling / [FoCo: una aplicación Shiny para la evaluación formativa usando modelos de diagnóstico cognitivo]

Combining formative and summative evaluations could improve assessment. Cognitive diagnosis modeling (CDM) has been proposed as a tool for diagnosing students’ strengths and weaknesses in formative assessment. However, there is no user-friendly software to implement it. For this reason, a Shiny app, FoCo, has been developed (https://foco.shinyapps.io/FoCo/), to conduct CDM and classical test theory analyses. The responses from 86 undergraduate students to a research methods course examination were analyzed. Students’ strengths and needs were diagnosed concerning their dominance of the syllabus contents and the first three competencies in Bloom’s taxonomy. The validity of the results was analyzed. The exam showed acceptable about evaluating students’ knowledge, as students with similar scores showed different strengths and weaknesses. Additionally, these attributes were found to predict different relevant criteria. It is expected that FoCo’s easiness to use promotes the employment of CDM in real educational settings.(AU)

La combinación de evaluaciones formativas y sumativas podría mejorar la evaluación. El modelado de diagnóstico cognitivo (MDC) se ha propuesto para diagnosticar fortalezas y debilidades de estudiantes en la evaluación formativa. Sin embargo, ningún software permite implementarlo fácilmente. Así, se ha desarrollado FoCo (https://foco.shinyapps.io/FoCo/), permitiendo realizar análisis MDC y teoría clásica de tests. Se analizaron respuestas de 86 estudiantes de grado a un examen de métodos de investigación, diagnosticándose sus fortalezas y necesidades en cuanto a su dominio de los contenidos de la asignatura y las tres primeras competencias de la taxonomía de Bloom y se analizó la validez de los resultados. El análisis ha sido informativo, ya que para estudiantes con puntuaciones similares ha sido posible detectar diferentes fortalezas y debilidades. Además, se encontró que estos atributos predicen criterios relevantes. Se espera que FoCo facilite el uso de MDC en contextos educativos.(AU)

A Modern Computer Application to Model Rare Earth Element Ion Behavior in Adsorptive Membranes and Materials.

The following paper offers a modern REE 1.0 computer application designed to model the behavior of REE ions in adsorptive materials and membranes. The current version of the application is based on several models, such as the Lagergren pseudo-first order, pseudo-second-order and Elovich kinetic models, and the intraparticle diffusion model, the diffusion-chemisorption model, and the Boyd model. The application has been verified on a sample of four different types of adsorptive materials and membranes. The proposed application allowed the analysis of kinetics, but also the mechanisms of the adsorption process, especially those responsible for the rate-determining steps. It was found that Lagergren pseudo-second-order kinetic model was the best-fit model to describe the adsorption behavior of REE ions onto the novel materials and membranes. Other models determined the process of chemisorption was in force for the analyzed cases, and the mechanisms controlling the adsorption processes are diffusion-chemisorption and adsorption is mostly controlled by film diffusion. Additionally, characteristic parameters, such as qe designated from two different models, showed very similar values, which indicates the correctness of the analysis.

Application of Chaos Mutation Adaptive Sparrow Search Algorithm in Edge Data Compression.

In view of the large amount of data collected by an edge server, when compression technology is used for data compression, data classification accuracy is reduced and data loss is large. This paper proposes a data compression algorithm based on the chaotic mutation adaptive sparrow search algorithm (CMASSA). Constructing a new fitness function, CMASSA optimizes the hyperparameters of the Convolutional Auto-Encoder Network (CAEN) on the cloud service center, aiming to obtain the optimal CAEN model. The model is sent to the edge server to compress the data at the lower level of edge computing. The effectiveness of CMASSA performance is tested on ten high-dimensional benchmark functions, and the results show that CMASSA outperforms other comparison algorithms. Subsequently, experiments are compared with other literature on the Multi-class Weather Dataset (MWD). Experiments show that under the premise of ensuring a certain compression ratio, the proposed algorithm not only has better accuracy in classification tasks than other algorithms but also maintains a high degree of data reconstruction.

mPD: Introduction of a single player prisoner's dilemma application for medical experiments.

BACKGROUND: Prisoner's dilemma is one of the most popular concepts among scientific literature. In medical literature the majority of prisoner's dilemma experiments with human participants implement computerized means. Despite this, there is no shared validated tool for prisoner's dilemma tasks. METHODS: The application is developed in Javascript programming language and makes use of the pixijs library for WebGL rendering. To create a custom trial, a set of variables have to be set. These refer to the linguistics, user controls, available choices, computer strategy, interaction flow (simultaneous or sequential), opponent's choice prediction requirement, noise induction, human opponent behavior simulation, the way resulted data will be handled and more. Results are in JSON format and include time interval data. RESULTS: We have developed an application which, given the various parameters experimenter can modify, is able to simulate a large number of single player prisoner's dilemma versions. It is open source with no installation requirement, executable by any modern internet browser remotely or locally with the ability to post data results either locally or remotely. Experimenter only has to modify certain starting values in order to create his design of choice. Two examples are included, with initial settings and results, to demonstrate the use and validity of the application. CONCLUSIONS: Our aim is to assist future researchers in their methodological designs. In this scope, our application, has the minimum requirements, can be served either locally or remotely, has a wide range of modifiable parameters and takes care of the resulted data. In the long term, a shared and validated tool would contribute to increasing methodologies' credibility and mitigating cross-validation discrepancies.

Efficacy of Superselective Conventional Transarterial Chemoembolization Using Guidance Software for Hepatocellular Carcinoma within Three Lesions Smaller Than 3 cm.

The indication of transarterial chemoembolization (TACE) has advanced to hepatocellular carcinoma (HCC) of Barcelona Clinic Liver Cancer (BCLC) stage A when surgical resection (SR), thermal ablation, and bridging to transplantation are contraindicated; however, TACE for small HCC is frequently difficult and ineffective because of less hypervascularity and the presence of tumor portions receiving a dual blood supply. Here, we report outcomes of superselective conventional TACE (cTACE) for 259 patients with HCCs within three lesions smaller than 3 cm using guidance software. Automated tumor feeder detection (AFD) functionality was applied to identify tumor feeders on cone-beam computed tomography during hepatic arteriography (CBCTHA) data. When it failed, the feeder was identified by manual feeder detection functionality and/or selective angiography and CBCTHA. Regarding the technical success in 382 tumors (mean diameter, 17.2 ± 5.9 mm), 310 (81.2%) were completely embolized with a safety margin (5 mm wide for HCC ≤25 mm and 10 mm wide for HCC >25 mm). In 61 (16.0%), the entire tumor was embolized but the safety margin was not uniformly obtained. The entire tumor was not embolized in 11 (2.9%). Regarding the tumor response at 2-3 months after cTACE in 303 tumors excluding those treated with combined radiofrequency ablation (RFA) or SR and lost to follow-up, 287 (94.7%) were classified into complete response, seven (2.3%) into partial response, and nine (3.0%) into stable disease. The mean follow-up period was 44.9 ± 27.6 months (range, 1-109) and the cumulative local tumor progression rates at 1, 3, 5, and 7 years were 17.8, 27.8, 32.0, and 36.0%, respectively. The 1-, 3-, 5-, and 7-year overall and recurrence-free survival rates in 175 patients, excluding those with Child-Pugh C class, who died of other malignancies, or who underwent combined RFA or hepatic resection, were 97.1 and 68.7, 82.8 and 34.9, 64.8 and 20.2, and 45.3 and 17.3%, respectively. Our results indicate the efficacy of superselective cTACE using guidance software for HCC within three lesions smaller than 3 cm.

Digital Integration and Automated Assessment of Eye-Tracking and Emotional Response Data Using the BioSensory App to Maximize Packaging Label Analysis.

New and emerging non-invasive digital tools, such as eye-tracking, facial expression and physiological biometrics, have been implemented to extract more objective sensory responses by panelists from packaging and, specifically, labels. However, integrating these technologies from different company providers and software for data acquisition and analysis makes their practical application difficult for research and the industry. This study proposed a prototype integration between eye tracking and emotional biometrics using the BioSensory computer application for three sample labels: Stevia, Potato chips, and Spaghetti. Multivariate data analyses are presented, showing the integrative analysis approach of the proposed prototype system. Further studies can be conducted with this system and integrating other biometrics available, such as physiological response with heart rate, blood, pressure, and temperature changes analyzed while focusing on different label components or packaging features. By maximizing data extraction from various components of packaging and labels, smart predictive systems can also be implemented, such as machine learning to assess liking and other parameters of interest from the whole package and specific components.

Classification of Myocardial 18F-FDG PET Uptake Patterns Using Deep Learning.

PURPOSE: To perform automated myocardial segmentation and uptake classification from whole-body fluorine 18 fluorodeoxyglucose (FDG) PET. MATERIALS AND METHODS: In this retrospective study, consecutive patients who underwent FDG PET imaging for oncologic indications were included (July-August 2018). The left ventricle (LV) on whole-body FDG PET images was manually segmented and classified as showing no myocardial uptake, diffuse uptake, or partial uptake. A total of 609 patients (mean age, 64 years ± 14 [standard deviation]; 309 women) were included and split between training (60%, 365 patients), validation (20%, 122 patients), and testing (20%, 122 patients) datasets. Two sequential neural networks were developed to automatically segment the LV and classify the myocardial uptake pattern using segmentation and classification training data provided by human experts. Linear regression was performed to correlate findings from human experts and deep learning. Classification performance was evaluated using receiver operating characteristic (ROC) analysis. RESULTS: There was moderate agreement of uptake pattern between experts and deep learning (as a fraction of correctly categorized images) with 78% (36 of 46) for no uptake, 71% (34 of 48) for diffuse uptake, and 71% (20 of 28) for partial uptake. There was no bias in LV volume for partial or diffuse uptake categories (P = .56); however, deep learning underestimated LV volumes in the no uptake category. There was good correlation for LV volume (R 2 = 0.35, b = .71). ROC analysis showed the area under the curve for classifying no uptake and diffuse uptake was high (> 0.90) but lower for partial uptake (0.77). The feasibility of a myocardial uptake index (MUI) for quantifying the degree of myocardial activity patterns was shown, and there was excellent visual agreement between MUI and uptake patterns. CONCLUSION: Deep learning was able to segment and classify myocardial uptake patterns on FDG PET images.Keywords: PET, Heart, Computer Aided Diagnosis, Computer Application-Detection/DiagnosisSupplemental material is available for this article.©RSNA, 2021.

GO Enrichment Analysis for Differential Proteomics Using ProteoRE.

With the increased simplicity of producing proteomics data, the bottleneck has now shifted to the functional analysis of large lists of proteins to translate this primary level of information into meaningful biological knowledge. Tools implementing such approach are a powerful way to gain biological insights related to their samples, provided that biologists/clinicians have access to computational solutions even when they have little programming experience or bioinformatics support. To achieve this goal, we designed ProteoRE (Proteomics Research Environment), a unified online research service that provides end-users with a set of tools to interpret their proteomics data in a collaborative and reproducible manner. ProteoRE is built upon the Galaxy framework, a workflow system allowing for data and analysis persistence, and providing user interfaces to facilitate the interaction with tools dedicated to the functional and the visual analysis of proteomics datasets. A set of tools relying on computational methods selected for their complementarity in terms of functional analysis was developed and made accessible via the ProteoRE web portal. In this chapter, a step-by-step protocol linking these tools is designed to perform a functional annotation and GO-based enrichment analyses applied to a set of differentially expressed proteins as a use case. Analytical practices, guidelines as well as tips related to this strategy are also provided. Tools, datasets, and results are freely available at http://www.proteore.org , allowing researchers to reuse them.

Integrating old and new complexity measures toward automated seizure detection from long-term video EEG recordings.

Automated seizure detection in long-term video-EEG recordings is far from being integrated into common clinical practice. Here, we leverage classical and state-of-the-art complexity measures to robustly and automatically detect seizures from scalp recordings. Brain activity is scored through eight features, encompassing traditional time domain and novel measures of recurrence. A binary classification algorithm tailored to treat unbalanced dataset is used to determine whether a time window is ictal or non-ictal from its features. The application of the algorithm on a cohort of ten adult patients with focal refractory epilepsy indicates sensitivity, specificity, and accuracy of 90%, along with a true alarm rate of 95% and less than four false alarms per day. The proposed approach emphasizes ictal patterns against noisy background without the need of data preprocessing. Finally, we benchmark our approach against previous studies on two publicly available datasets, demonstrating the good performance of our algorithm.

Pressure distribution analysis of X-Ray table mattresses.

BACKGROUND: There is a risk of developing pressure ulcers from lying on an X-ray table mattress, if the mattress pressure redistribution properties are poor. AIM: To assess the pressure redistribution properties of 'new' and 'in current clinical use' X-ray table mattresses. METHODS AND MATERIALS: Twenty one X-ray table mattresses, each of 2.5 cm thickness, were evaluated. An anthropomorphic human phantom of adult stature with five different weights (minimum, first quartile, mean, third quartile and maximum) was used to simulate human head, pelvis and heels (pressure ulcer jeopardy areas). Using Xsensor technology, peak pressure was measured and Interface Pressure Ratio was calculated for the three pressure ulcer jeopardy areas 'with' and 'without' an X-ray table mattress. RESULTS: For all mattresses, statistically significant differences (p < 0.05) were found between the peak pressure values with and without using an X-ray table mattress for the three pressure ulcer jeopardy areas; similarly, for all mattresses, statistically significant differences (p < 0.05) were found between the Interface Pressure Ratio values with and without using x-ray table mattress. The type and age of the mattress was observed to have an impact on peak pressure values and Interface Pressure Ratios, with older mattresses performing worse. CONCLUSION: Peak pressure values and Interface Pressure Ratios are reduced significantly when using newer X-ray table mattresses. This could be because newer mattresses use more appropriate materials in their construction and/or older mattresses have lost their pressure redistribution properties. Radiology departments should consider assessing mattresses pressure redistribution properties, perhaps on an annual basis.

Segmentation procedures for the assessment of paranasal sinuses volumes.

BACKGROUND: The paranasal sinuses are complex anatomical structures, characterised by highly variable shape, morphology and size. With the introduction of multidetector scanners and the development of many post-processing possibilities, computed tomography became the gold standard technique to image the paranasal sinuses. Segmentation allows the extraction of metrical and shape data of these anatomical components that can be applied for diagnostic, education, surgical planning and simulation, and to plan minimally invasive interventions in otorhinolaryngology and neurosurgery. DISCUSSION: Our aim was to provide a review of the existing literature on segmentation, its types and application, and the data obtained from this procedure. The literature search was conducted on PubMed (including Medline), ScienceDirect and Google Scholar databases, using the keywords as follows: 'paranasal sinuses', 'frontal sinus', 'maxillary sinus', 'sphenoid sinus', 'ethmoid sinus', in all possible combinations with the keywords 'segmentation' and 'volumetric analysis'. Inclusion criteria were: articles written in English, on living human subjects, on the adult population and focused on paranasal sinuses analysis. CONCLUSION: This article provides an overview of the types and main application of segmentation procedures on paranasal sinuses, and the results provided by the studies on this topic.

Pesquisador Virtual: solución informática para la detección de casos sospechosos de COVID-19 / Pesquisador Virtual: IT solution for the detection of suspected COVID-19 cases

Se presenta el Pesquisador Virtual, solución informática desarrollada por la Universidad de las Ciencias Informáticas (UCI) en colaboración con el Ministerio de Salud Pública (MINSAP), que mediante encuestas permite captar información del estado de salud (síntomas y padecimientos) de la población, como complemento al proceso de pesquisa activa, en el marco del enfrentamiento epidemiológico a la pandemia COVID-19. Para la realización de la encuesta se cuenta con una solución móvil y una solución Web. El resultado de la encuesta (información estadística y nominal) es monitorizada en tiempo real mediante gráficos y estadísticas por las diferentes instancias del MINSAP, tanto a nivel nacional como provincial y municipal, mostrando reportes personalizados según el nivel de acceso en dependencia del nivel del organismo. Con esta información, diferentes instancias del sistema de Salud actúan de manera inmediata, sobre todo la atención primaria de salud (APS), para evitar la propagación de la epidemia y la atención temprana de posibles contagiados(AU)

This article presents Pesquisador Virtual, a computer solution developed by the University of Informatics Sciences (UCI) in collaboration with the Ministry of Public Health (MINSAP), which, through surveys, allows the capture of information about the population's health status, as a complement to the active inquiry process in the context of the epidemiological response to the COVID-19 pandemic. A mobile and Web solutions are available for the presentation of the survey. The results of the survey (statistical and nominal information) are monitored in real time through graphics and statistics by the different MINSAP structures at the national, provincial and municipal levels, with variable access depending on the level of the structures. With this information, different departments of the health system act immediately, especially Primary Health Care (APS), to prevent the spread of the epidemic and to provide early care for those who may be infected(AU)

Evaluation of Functional Abilities in 0-6 Year Olds: an Analysis with the eEarlyCare Computer Application.

The application of Industry 4.0 to the field of Health Sciences facilitates precise diagnosis and therapy determination. In particular, its effectiveness has been proven in the development of personalized therapeutic intervention programs. The objectives of this study were (1) to develop a computer application that allows the recording of the observational assessment of users aged 0-6 years old with impairment in functional areas and (2) to assess the effectiveness of computer application. We worked with a sample of 22 users with different degrees of cognitive disability at ages 0-6. The eEarlyCare computer application was developed with the aim of allowing the recording of the results of an evaluation of functional abilities and the interpretation of the results by a comparison with "normal development". In addition, the Machine Learning techniques of supervised and unsupervised learning were applied. The most relevant functional areas were predicted. Furthermore, three clusters of functional development were found. These did not always correspond to the disability degree. These data were visualized with distance map techniques. The use of computer applications together with Machine Learning techniques was shown to facilitate accurate diagnosis and therapeutic intervention. Future studies will address research in other user cohorts and expand the functionality of their application to personalized therapeutic programs.

CardiacPBPK: A tool for the prediction and visualization of time-concentration profiles of drugs in heart tissue.

BACKGROUND AND OBJECTIVE: Prediction of drug concentration in heart tissue is important in terms of drug safety and efficacy. This work presents the Open-Source CardiacPBPK platform for the prediction of the time-concentration profile of drugs, which could potentially reduce the risk of drug development failure due to cardiotoxicity. The objective of the CardiacPBPK development is to accelerate and simplify the in-silico toxicological assessment of new drugs, and to provide supportive material for the research community to use. METHODS: The CardiacPBPK software provides a modular implementation of the PBPK model of heart tissue. It can be easily accessed via the Internet or installed locally. The graphical user interface and tabular design are easy to configure and use. RESULTS: CardiacPBPK is a tool designed to predict and visualize the time-concentration profiles of a parent compound, and one metabolite, in venous plasma and heart tissue after oral or intravenous drug administration. CardiacPBPK is built on the R-environment framework and supports shiny application features such as interactive visualization of the results, and web applications interface by default. A shiny application refers to a computer program created with the use of shiny package in R. The application is freely available at https://github.com/jszlek/CardiacPBPK and https://sourceforge.net/projects/cardiacpbpk/. This open-source application runs on all platforms supporting R-environment (Linux, Windows, Mac OS X, Solaris). CONCLUSIONS: We demonstrate the application of CardiacPBPK by simulating the study of amitriptyline intoxication in the case of CYP2D6 genetic polymorphism.

Bioinformatics Tools and Workflow to Select Blood Biomarkers for Early Cancer Diagnosis: An Application to Pancreatic Cancer.

Secretome proteomics for the discovery of cancer biomarkers holds great potential to improve early cancer diagnosis. A knowledge-based approach relying on mechanistic criteria related to the type of cancer should help to identify candidates from available "omics" information. With the aim of accelerating the discovery process for novel biomarkers, a set of tools is developed and made available via a Galaxy-based instance to assist end-users biologists. These implemented tools proceed by a step-by-step strategy to mine transcriptomics and proteomics databases for information relating to tissue specificity, allow the selection of proteins that are part of the secretome, and combine this information with proteomics datasets to rank the most promising candidate biomarkers for early cancer diagnosis. Using pancreatic cancer as a case study, this strategy produces a list of 24 candidate biomarkers suitable for experimental assessment by MS-based proteomics. Among these proteins, three (SYCN, REG1B, and PRSS2) were previously reported as circulating candidate biomarkers of pancreatic cancer. Here, further refinement of this list allows to prioritize 14 candidate biomarkers along with their associated proteotypic peptides for further investigation, using targeted MS-based proteomics. The bioinformatics tools and the workflow implementing this strategy for the selection of candidate biomarkers are freely accessible at http://www.proteore.org.

New Tools in Education: Development and Learning Effectiveness of a Computer Application for Use in a University Biology Curriculum.

In recent years student exposure to computer applications has increased at an unprecedented rate. Yet the use of these promising technologies in education remains in its infancy. The growing practice of 'gamification' offers today's educators the means of conveying their lessons in a more engaging way, by utilising computer game mechanics. However, many of these learning tools have not been empirically evaluated. This research investigated the development of a desktop computer application, to replace an existing learning resource, a video, currently used by over 700 life sciences students a year in one of the top 100 universities of the world. The interactive game presents the same essential information as the video, on key anatomical features of mammalian skulls, and provides student self-testing. Results from a two-treatment, pre- and post-intervention experimental design suggest the new product is better for providing both knowledge acquisition and a positive learning experience. Nevertheless, the results are unlikely to be statistically significant. Insights from the findings are discussed and directions for future research are given.

Automated spectrographic seizure detection using convolutional neural networks.

PURPOSE: Non-convulsive seizures are common in critically ill patients, and delays in diagnosis contribute to increased morbidity and mortality. Many intensive care units employ continuous EEG (cEEG) for seizure monitoring. Although cEEG is continuously recorded, it is often reviewed intermittently, which may delay seizure diagnosis and treatment. This may be mitigated with automated seizure detection. In this study, we develop and evaluate convolutional neural networks (CNN) to automate seizure detection on EEG spectrograms. METHODS: Adult EEGs (12 patients, 12 EEGs, 33 seizures) from New-York Presbyterian Hospital (NYP) and pediatric EEGs (22 patients, 130 EEGs, 177 seizures) from Children's Hospital Boston (CHB) were converted into spectrograms. To simulate a telemetry display, seizure and non-seizure events on spectrograms were sequentially sampled as images across a detection window (26,380 total images). Four CNN models of increasing complexity (number of layers) were trained, cross-validated, and tested on CHB and NYP spectrographic images. All CNNs were based on the VGG-net architecture, with adjustments to alleviate overfitting. RESULTS: For spectrographically visible seizures, two CNN models (containing 4 and 7 convolution layers) achieved >90% seizure detection sensitivity and specificity on the CHB test set and >90% sensitivity and 75-80% specificity on the NYP test set. The one CNN model (10 convolution layers) did not converge during training; while another CNN (2 convolution layers) performed poorly (60% sensitivity and 32% specificity) on the NYP test set. CONCLUSIONS: Seizure detection on EEG spectrograms with CNN models is feasible with sensitivity and specificity potentially suitable for clinical use.

Designing an In Silico Strategy to Select Tissue-Leakage Biomarkers Using the Galaxy Framework.

Knowledge-based approaches using large-scale biological ("omics") data are a powerful way to identify mechanistic biomarkers, provided that scientists have access to computational solutions even when they have little programming experience or bioinformatics support. To achieve this goal, we designed a set of tools under the Galaxy framework to allow biologists to define their own strategy for reproducible biomarker selection. These tools rely on retrieving experimental data from public databases, and applying successive filters derived from information relating to disease pathophysiology. A step-by-step protocol linking these tools was implemented to select tissue-leakage biomarker candidates of myocardial infarction. A list of 24 candidates suitable for experimental assessment by MS-based proteomics is proposed. These tools have been made publicly available at http://www.proteore.org , allowing researchers to reuse them in their quest for biomarker discovery.