Computational Science – ICCS 2024

This study introduces an innovative memetic algorithm for optimizing the consensus of well-adapted techniques for the inference of gene regulation networks. Building on the methodology of a previous proposal (GENECI), this research adds a local search phase that incorporates prior knowledge about gene interactions, thereby enhancing the optimization process under the influence of domain expert. The algorithm focuses on the evaluation of candidate solutions through a detailed evolutionary process, where known gene interactions guide the evolution of such solutions (individuals). This approach was subjected to rigorous testing using benchmarks from editions 3 and 4 of the DREAM challenges and the yeast network of IRMA, demonstrating a significant improvement in accuracy compared to previous related approaches. The results highlight the effectiveness of the algorithm, even when only 5% of the known interactions are used as a reference. This advancement represents a significant step in the inference of gene regulation networks, providing a more precise and adaptable tool for genomic research.

The paper proposes a method of human sex recognition using individual gait features extracted by measures describing the dimensionality and uncertainty of non-linear dynamical systems. The correlation dimension and sample entropy are computed for time series representing angles of skeletal body joints as well as whole-body orientation and translation. Two aggregation strategies for pose parameters are used – averaging of Euler angles triplets and taking an angle of 3D rotation. In the baseline variant, the distinction between females and males is performed by thresholding the obtained measure values. Moreover, the supervised classification is carried out for the complex gait descriptors characterizing the movements of all bone segments. In the validation experiments, highly precise motion capture measurements containing data of 25 female and 30 male individuals are used. The obtained, at least promising, performance assessed by correct classification rate, the area under the receiver operating characteristic curve, and average precision, is higher than 89%, 96%, and 96%, respectively, and exceeds our expectations. Moreover, the classification accuracy based on a ranking of skeletal joints, as well as whole-body orientation and translation evaluating sex-discriminative traits incorporated in the movements of bone segments, is formed.

In this paper a multi-domain multi-task algorithm for feature selection in bulk RNAseq data is proposed. Two datasets are investigated arising from mouse host immune response to Salmonella infection. Data is collected from several strains of collaborative cross mice. Samples from the spleen and liver serve as the two domains. Several machine learning experiments are conducted and the small subset of discriminative across domains features have been extracted in each case. The algorithm proves viable and underlines the benefits of across domain feature selection by extracting new subset of discriminative features which couldn’t be extracted only by one-domain approach.

Parkinson’s disease (PD) is a neurological disorder defined by the gradual loss of dopaminergic neurons in the substantia nigra pars compacta, which causes both motor and non-motor symptoms. Understanding the neuronal processes that underlie PD is critical for creating successful therapies. This work presents a novel strategy that combines machine learning (ML) and stochastic modelling with connectomic data to understand better the complicated brain pathways involved in PD pathogenesis. We use modern computational methods to study large-scale neural networks to identify neuronal activity patterns related to PD development. We aim to define the subtle structural and functional connection changes in PD brains by combining connectomic with stochastic noises. Stochastic modelling approaches reflect brain dynamics’ intrinsic variability and unpredictability, shedding light on the origin and spread of pathogenic events in PD. We created a hybrid modelling formalism and a novel co-simulation approach to identify the effect of stochastic noises on the cortex-BG-thalamus (CBGTH) brain network model in a large-scale brain connectome. We use Human Connectome Project (HCP) data to elucidate a stochastic influence on the brain network model. Furthermore, we choose areas of the parameter space that reflect both healthy and Parkinsonian states and the impact of deep brain stimulation (DBS) on the subthalamic nucleus and thalamus. We infer that thalamus activity increases with stochastic disturbances, even in the presence of DBS. We predicted that lowering the effect of stochastic noises would increase the healthy state of the brain. This work aims to unravel PD’s complicated neuronal activity dynamics, opening up new options for therapeutic intervention and tailored therapy.

Artificial intelligence has contributed to advancements across various industries. However, the rapid growth of artificial intelligence technologies also raises concerns about their environmental impact, due to associated carbon footprints to train computational models. Fetal brain segmentation in medical imaging is challenging due to the small size of the fetal brain and the limited image quality of fast 2D sequences. Deep neural networks are a promising method to overcome this challenge. In this context, the construction of larger models requires extensive data and computing power, leading to high energy consumption. Our study aims to explore model architectures and compression techniques that promote energy efficiency by optimizing the trade-off between accuracy and energy consumption through various strategies such as lightweight network design, architecture search, and optimized distributed training tools. We have identified several effective strategies including optimization of data loading, modern optimizers, distributed training strategy implementation, and reduced floating point operations precision usage with light model architectures while tuning parameters according to available computer resources. Our findings demonstrate that these methods lead to satisfactory model performance with the low energy consumption during deep neural network training for medical image segmentation.

Negation detection refers to the automatic identification of linguistic expression that convey negation within a textual content. In medical and biomedical context, the negation detection plays a pivotal role in understanding clinical documentation and extracting meaningful insights. In this paper, we survey 16 articles published from 2005 to 2023 and focusing on negation detection within medical domain. Our evaluation framework encompass both methodological aspects and application-oriented considerations. Specifically, we discuss the used approaches, the employed methodology, the specific tasks addressed, the target language of textual analysis, and the evaluation metrics used. On the application front, for each reviewed study, we delineate the medical domains under investigation (e.g., cardiology, oncology), the types of data analyzed, and the availability of datasets. The majority of reviewed works are conducted in English, with a prevalence of machine learning and deep learning approaches, and classic classification evaluation metrics. Application domains exhibit heterogeneity, with a slight predominance in oncology, and diverse data sources including EHRs, abstracts, scientific papers, and web-derived information (e.g., Wikipedia or blog entries). Throughout this review, we will identify limitations and gaps in this research area, as well as examine the benefits it could bring to the scientific community and the methods currently employed.

The development of more complex biomarker selection protocols based on the machine learning (ML) approach, with additional processing of information from biological databases (DB), is important for the accelerated development of molecular diagnostics and therapy.In this study, we present EnsembleFS user-friendly R toolkit (R package and Shiny web application) for heterogeneous ensemble feature selection (EFS) of molecular omics data that also supports users in the analysis and interpretation of the most relevant biomarkers. EnsembleFS is based on five feature filters (FF), namely, U-test, minimum redundancy maximum relevance (MRMR), Monte Carlo feature selection (MCFS), and multidimensional feature selection (MDFS) in 1D and 2D versions. It uses supervised ML methods to evaluate the quality of the set of selected features and retrieves the biological characteristics of biomarkers online from the nine DB, such as Gene Ontology, WikiPathways, and Human Protein Atlas. The functional modules to identify potential candidate biomarkers, evaluation, comparison, analysis, and visualization of model results make EnsembleFS a useful tool for selection, random forest (RF) binary classification, and comprehensive biomarker analysis.

The analysis of Disease-Disease Associations (DDA) and Gene-Disease Associations (GDA) is a relevant task in bioinformatics. These are analysed to investigate the interactions between sets of diseases and genes as well as their similarity, e.g., to improve the phases of diagnosis, prognosis and treatment in medicine. Generally, the extraction of information of interest from large-scale data, usually heterogeneous and unstructured, is performed via time-consuming processes. Therefore, several computational approaches have been focused on their prediction through data integration and machine learning techniques.This paper presents a solution for Inferring DDA (IDDA) by integrating curated biomedical ontologies and medical dictionaries. It is able to extract a set of DDA using an in-house score based on the GDA. A preliminary step based on data enrichment retrieves the information about gene and disease, and it integrates these with a set of curated biological data ontologies and dictionaries. Specifically, IDDA extracts DDAs based on an in-house score, which uses GDAs for its evaluations. In a preliminary step, it performs data enrichment to retrieve concepts both for diseases and genes, by integrating several curated biomedical ontologies and medical dictionaries.

We propose a new graphical model to describe the comorbidity of allergic diseases. We present our model in two versions. First, we introduce a generative model that reflects the variables’ causal relationships. Then, we propose an approximation of the generative model by a misspecified model, which is computationally more efficient and easily interpretable. In both versions of our model, we consider typical allergic disease symptoms and covariates. We consider two directed acyclic graphs (DAGs). The first one describes information about the coexistence of certain allergic diseases (binary variables). The second graph describes the relationships between particular symptoms and the occurrence of these diseases. In the generative model, the edges lead from diseases to symptoms, corresponding to causal relations. In the misspecified model, we reverse the direction of edges: they lead from symptoms to diseases. The proposed model is evaluated on a cross-sectional multicentre study in Poland ( www.ecap.pl ). An assessment of the stability of the proposed model is obtained using the bootstrap and jackknife techniques. Our results show that the misspecified model is a good approximation of the generative model and helps predict the incidence of allergic diseases.

Transmembrane proteins (TMPs) are crucial to cell biology, making up about 30% of all proteins based on genomic data. Despite their importance, most of the available software for aligning protein sequences focuses on soluble proteins, leaving a gap in tools specifically designed for TMPs. Only a few methods target TMP alignment, with just a couple of the available to researchers. Considering that there are a few particular differences that ought to be taken into consideration aligning TMPs sequences, standard MSA methods are ineffective to align TMPs. In this paper, we present TM-MSAligner, a software tool designed to deal with the multiple sequence alignment of TMPs by using a multi-objective evolutionary algorithm. Our software include features such as transmembrane substitution matrix dynamically used according to the topology region, a high penalty to gap opening and extending, and two MSA quality scores, Sum-Of-Pairs with Topology Prediction and Aligned Segments, that can be optimized at the same time. This approach reduce the number of Transmembrane (TM) and non-Transmembrane (non-TM) broken regions and improve the TMP quality score. TM-MSAligner outputs the results in an HTML format, providing an interactive way for users to visualize and analyze the alignment. This feature allows for the easy identification of each topological region within the alignment, facilitating a quicker and more effective analysis process for researchers.

The study of the relationship between brain neuron activity and behavioral responses of humans and other animals is an area of interest, although it has received relatively little attention from scientific biology and medical research centers. In this paper, we consider the problem of determining a mouse position in a circular track based on its neural activity data, and investigate the use of machine learning for solving this problem. The study is conducted in two parts: a classification task, where the model predicts which sector of the track the mouse is in at a particular time, and a regression task, where it predicts exact coordinates for each time step. We propose a neural network-based solution for both tasks, based on a graph of brain neuron activity. Accuracy results were obtained: 89% for classification and 93% for regression.

Since the launch of various generative AI tools, scientists have been striving to evaluate their capabilities and contents, in the hope of establishing trust in their generative abilities. Regulations and guidelines are emerging to verify generated contents and identify novel uses. we aspire to demonstrate how ChatGPT claims are checked computationally using the rigor of network models. We aim to achieve fact-checking of the knowledge embedded in biological graphs that were contrived from ChatGPT contents at the aggregate level. We adopted a biological networks approach that enables the systematic interrogation of ChatGPT’s linked entities. We designed an ontology-driven fact-checking algorithm that compares biological graphs constructed from approximately 200,000 PubMed abstracts with counterparts constructed from a dataset generated using the ChatGPT-3.5 Turbo model. In 10-samples of 250 randomly selected records a ChatGPT dataset of 1000 “simulated” articles , the fact-checking link accuracy ranged from 70% to 86%. This study demonstrated high accuracy of aggregate disease-gene links relationships found in ChatGPT-generated texts.

Lipid microdomains are specialized structures that play crucial roles in various physiological and pathological processes, such as modulating immune responses, facilitating pathogen entry, and forming signaling platforms. In this study, we explored the dynamics and organization of lipid membranes using a combination of molecular dynamics simulations and a suite of machine learning (ML) techniques. Using ML algorithms, we accurately classified membrane regions into liquid order, liquid-disordered, or interfacial states, demonstrating the potential of computational methods to predict complex biological organizations. Our investigation was mainly focused on two lipid systems: POPC/PSM/CHOL, and DPPC/DLIPC/CHOL. The study underscores the dynamic interaction between ordered and disordered phases within cellular membranes, with a pivotal role of cholesterol in inducing domain formation.

Computational modeling emerged to address scientific problems by developing mathematical models for their description and creating computational codes to obtain solutions. Employing this technique in studying cardiac electrophysiology enables a better understanding of heart function, which requires considerable time and technological expertise. MonoWeb is a structured platform for simulating electrophysiological activity in cardiac tissues, using the monodomain model in a browser-based manner. This tool provides not only an accessible platform to simulate cardiac electrical activity but also integrates visualization and flexible configuration with an intuitive interface. Through communication with the MonoAlg3D simulator, it allows the input of advanced parameters, and different cellular models, including selecting arrhythmia examples, and stimuli, with the goal of making this experience easier and practical for electrophysiology professionals.

This study introduces a novel Convolutional Neural Network (CNN) approach for breast cancer diagnosis, which seamlessly integrates segmentation and classification. The segmentation process achieves high precision, with Jaccard Index (JI) values of 0.89, 0.92, and 0.87 for Normal, Benign, and Malignant regions, respectively, resulting in an overall JI of 0.896. Similarly, the Dice Similarity Coefficient (DSC) values are notably high, with 0.94, 0.96, and 0.92 for the corresponding regions, yielding an overall DSC of 0.943. The CNN model exhibits high accuracy, specificity, precision, recall, and F1 score across all classes, establishing its reliability for clinical applications. This research comprehensively evaluates the model’s performance metrics, addressing challenges in breast cancer diagnostics and proposing an innovative CNN-based solution. Beyond immediate applications, it lays a robust foundation for future medical imaging advancements, enhancing diagnostic accuracy and patient outcomes.

Image recognition techniques heavily rely on abundant labeled data, particularly in medical contexts. Addressing the challenges associated with obtaining labeled data has led to the prominence of self-supervised learning and semi-supervised learning, especially in scenarios with limited annotated data. In this paper, we proposed an innovative approach by integrating self-supervised learning into semi-supervised models to enhance medical image recognition. Our methodology commences with pre-training on unlabeled data utilizing the BYOL method. Subsequently, we merge pseudo-labeled and labeled datasets to construct a neural network classifier, refining it through iterative fine-tuning. Experimental results on three different datasets demonstrate that our approach optimally leverages unlabeled data, outperforming existing methods in terms of accuracy for medical image recognition.

Using a minimal but sufficient closed-loop encapsulation and the theoretical framework of classical control, we implement and test the mathematical model of the baroregulation due to Mauro Ursino [24]. We present and compare data from a local relative sensitivity analysis and an input parameter orthogonality analysis from a regulated and then an equivalent unregulated cardiovascular model with a single ventricle and “CRC” Windkessel representation of the systemic circulation. We conclude: (i) a basic model of the closed-loop control is intrinsically stable; (ii) regulation generally (but not completely) suppresses the sensitivity of output responses on mechanical input parameters; (iii) with the sole exception of the regulation set-point, the mechanical input parameters are more influential on system outputs than the regulation input parameters. This work is the initial step for further analysis of more complex and computationally expensive models of the cardiovascular system, with baroreflex control, with possible applications in space-flight medicine or research on exercise intolerance.

The application of computational science methods and tools in healthcare is growing rapidly. These methods support decision-making and policy development. They are commonly used in decision support systems (DSSs) used in many fields. This paper presents a decision support system based on the newly developed SSP-SPOTIS (Strong Sustainable Paradigm based Stable Preference Ordering Towards Ideal Solution) method. The application of the proposed DSS is demonstrated in the example of assessing healthcare systems of selected countries concerning resilience to pandemic-type crisis phenomena. The developed method considers the strong sustainability paradigm by reducing linear compensation criteria with the possibility of its modeling. The research demonstrated the usefulness, reliability, and broad analytical opportunities of DSS based on SSP-SPOTIS in evaluation procedures focused on sustainability aspects considering a strong sustainability paradigm.

We present experimental data and modelling results investigating the effects of plasma-assisted air cleaning systems on reducing transmission of SARS-CoV-2 virus among pupils in a primary school in Amsterdam, the Netherlands. We equipped 4 classrooms (120 pupils) with the Novaerus NV800 ICU air cleaning system, and 8 classrooms (240 pupils) had standard ventilation systems. We found a significantly lower number of infections in classrooms with air cleaning systems in the first two weeks after instalment, suggesting that air cleaning decreases aerosol transmission. In the subsequent weeks, however, infection numbers increased in the Netherlands, and the difference between classrooms with and without air cleaning ceased to be significant. We analyzed the experimental results, performed a Kaplan-Meier survival estimation and developed a SIR-based computational model that simulates the results of this experiment. We performed sensitivity analysis, optimised model parameters, and tested several hypotheses. This research gives the potential for implementing improved air quality measures in public spaces, which could result in better air quality regulations in spaces such as schools.

Decision support systems in the medical domain is budding field that aims to improve healthcare and overall recovery for patients. While treatment remains specific to individual symptoms, the diagnosis of patients is fairly general. Interpreting the diagnosis and assigning the appropriate care treatment is a crucial part undertaken by medical professionals, however, in critical scenarios, having access to recommendations from a clinical decision support system may prove life-saving. We present a real-world application of inverse reinforcement learning (IRL) to assess the implicit cognitive state of doctors when evaluating decision support data on a patient’s risk of acquiring Type 2 Diabetes mellitus (T2DM). We show the underlying process of modelling a Markov Decision Process (MDP) using real-world clinical data and experiment with various policies extracted from sampled trajectories. The results provide insights into the approach to modelling real-world data into interpretable solutions via IRL.

This research investigates how collectives of doctors influence their diagnostic method preferences within small-world network social structures through participation in diverse types of medical practice-sharing activities across different scales. We propose an approach based on vectorization of the preferences for various diagnostic methods among physicians, quantifying their openness to these methods using the Shannon diversity index. Utilizing theoretical foundations from threshold models, influence models, and the Hegselmann-Krause model, we designed simulation experiments for teaching activities and seminars to explore the dynamic changes in preference vectors and Shannon diversity indices among these doctors in a small-world network. We evaluated our approach with a real-world data set on vertigo treatment by several clinical specialists of different specialty (neurologists, otolaryngologist). Building on real data from this initial group, we then simulated data for a large number of doctors from various medical communities to examine phenomena in larger-scale systems . Hierarchical networks featuring small-world properties were developed to simulate “local” within-community and “global” across-community seminars, reflecting different intra- and inter-community scenarios. The experiments show different patterns of practice converging during simulation in various scales and scenarios. The findings of this study provide significant insights for further research into practice-based knowledge sharing among healthcare professionals, highlighting the nuanced interplay between social network structures and professional consensus formation.

The German Federal Criminal Police Office (BKA) reported damages of 72.6 million euros due to billing fraud in the German healthcare system in 2022, an increase of 25% from the previous year. However, existing literature on automated healthcare fraud detection focuses on US, Taiwanese, or private data, and detection approaches based on individual claims are virtually nonexistent. In this work, we develop machine learning methods that detect fraud in German hospital billing data.The lack of publicly available and labeled datasets limits the development of such methods. Therefore, we simulated inpatient treatments based on publicly available statistics on main and secondary diagnoses, operations and demographic information. We injected different types of fraud that were identified from the literature. This is the first complete simulator for inpatient care data, enabling further research in inpatient care.We trained and compared several Machine Learning models on the simulated dataset. Gradient Boosting and Random Forest achieved the best results with a weighted F1 measure of approximately 80%. An in-depth analysis of the presented methods shows they excel at detecting compensation-related fraud, such as DRG upcoding. An impact analysis on private inpatient claims data of a big German health insurance company revealed that up to 12% of all treatments were identified as potentially fraudulent.

The aim of the research is to assess the applicability of methods of artificial intelligence to the analysis and prediction of infectious disease dynamics, with an aim to increase the speed of obtaining predictions along with enhancing quality of the results. To ensure the compliance of the forecasts with the natural laws governing the epidemic transmission, we employ Physics-Informed Neural Networks (PINN) as our main tool for the forecasting experiments. With the help of numerical experiments, we show the applicability of the approach to infectious disease modeling based on coupling classic approaches, namely, SIR models, and the cutting-edge research related to machine learning techniques. We compare the accuracy of different implementations of PINN along with the statistical models in the task of forecasting COVID incidence in Saint Petersburg, thus choosing the best modeling approach for this challenge. The results of the research could be incorporated into surveillance systems monitoring the advance of COVID and influenza incidence in Russian cities.

Chronic Kidney Disease (CKD) is a common disease with high incidence and high risk for the patients’ health when it degrades to its most advanced stages. When detected early, it is possible to slow down the progression of the disease, leading to an increased survival rate and lighter treatment. As a consequence, many prediction models have emerged for the prediction of CKD. However, few of them manage to efficiently predict the onset of the disease months to years prior. In this paper, we propose an artificial neural network combining the strengths of convolution and involution layers in order to predict the degradation of CKD to its later stages, based on a set of 25 common laboratory analyses as well as the age and gender of the patient. Using a dataset from a French medical laboratory containing more than 400 000 patients, we show that our model achieves better performance than state-of-the-art models, with a recall of 83%, F1-score of 76%, and 97% overall accuracy. The proposed method is flexible and easily applicable to other diseases, offering encouraging perspectives in the field of early disease prediction, as well as the use of involution layers for deep learning with time series.

Cervical cancer is the fourth most common cancer among women. Every year, more than 200,000 women die due to cervical cancer; however, it is a preventable disease if detected early. This study aims to detect cervical cancer by identifying the cytoplasm and nuclei from the background using deep learning techniques to automate the separation of a single cell. To preprocess the image, resizing and enhancement are adopted by adjusting the brightness and contrast of the image to remove noise in the image. The data is divided into 80% for training and 20% for testing to create models using deep neural networks. The U-Net serves as baseline network for image segmentation, with VGG19, ResNet50, MobileNet, EfficientNetB2 and DenseNet121 used as backbone. In cytoplasmic segmentation, EfficientNetB2 achieves a precision of 99.02%, while DenseNet121 reaches an accuracy of 98.59% for a single smear cell. For nuclei segmentation, EfficientNetB2 achieves an accuracy of 99.86%, surpassing ResNet50, which achieves 99.85%. As a result, deep learning-based image segmentation shows promising result in separating the cytoplasm and nuclei from the background to detect cervical cancer. This is helpful for cytotechnicians in diagnosis and decision-making.

Machine learning on large-scale genomic or transcriptomic data is important for many novel health applications. For example, precision medicine tailors medical treatments to patients on the basis of individual biomarkers, cellular and molecular states, etc. However, the data required is sensitive, voluminous, heterogeneous, and typically distributed across locations where dedicated machine learning hardware is not available. Due to privacy and regulatory reasons, it is also problematic to aggregate all data at a trusted third party. Federated learning is a promising solution to this dilemma, because it enables decentralized, collaborative machine learning without exchanging raw data.In this paper, we perform comparative experiments with the federated learning frameworks TensorFlow Federated and Flower. Our test case is the training of disease prognosis and cell type classification models. We train the models with distributed transcriptomic data, considering both data heterogeneity and architectural heterogeneity. We measure model quality, robustness against privacy-enhancing noise, computational performance and resource overhead. Each of the federated learning frameworks has different strengths. However, our experiments confirm that both frameworks can readily build models on transcriptomic data, without transferring personal raw data to a third party with abundant computational resources.

This work introduces an enhanced methodology in the domain of eXplainable Artificial Intelligence (XAI) for visualizing local explanations of black-box, feature-based models, such as LIME and SHAP, enabling both domain experts and non-specialists to identify the segments of Time Series (TS) data that are significant for machine learning model interpretations across classes. By applying this methodology to electrocardiogram (ECG) data for anomaly detection, distinguishing between healthy and abnormal segments, we demonstrate its applicability not only in healthcare diagnostics but also in predictive maintenance scenarios. Central to our contribution is the development of the AUC Perturbational Accuracy Loss metric (AUC-PALM), which facilitates the comparison of explainer fidelity across different models. We advance the field by evaluating various perturbation methods, demonstrating that perturbations centered on time series prototypes and those proportional to feature importance outperform others by offering a more distinct comparison of explainer fidelity with the underlying black-box model. This work lays the groundwork for broader application and understanding of XAI in critical decision-making processes.

The development of black-box survival models has created a need for methods that explain their outputs, just as in the case of traditional machine learning methods. Survival models usually predict functions rather than point estimates. This special nature of their output makes it more difficult to explain their operation. We propose a method to generate plausible counterfactual explanations for survival models. The method supports two options that handle the special nature of survival models’ output. One option relies on the Survival Scores, which are based on the area under the survival function, which is more suitable for proportional hazard models. The other one relies on Survival Patterns in the predictions of the survival model, which represent groups that are significantly different from the survival perspective. This guarantees an intuitive well-defined change from one risk group (Survival Pattern) to another and can handle more realistic cases where the proportional hazard assumption does not hold. The method uses a Particle Swarm Optimization algorithm to optimize a loss function to achieve four objectives: the desired change in the target, proximity to the explained example, likelihood, and the actionability of the counterfactual example. Two predictive maintenance datasets and one medical dataset are used to illustrate the results in different settings. The results show that our method produces plausible counterfactuals, which increase the understanding of black-box survival models.

In this research, we investigate the effectiveness of Large Language Models (LLMs) in answering health-related questions. The rapid growth and adoption of LLMs, such as ChatGPT, have raised concerns about their accuracy and robustness in critical domains such as Health Care and Medicine. We conduct a comprehensive study comparing multiple LLMs, including recent models like GPT-4 or Llama2, on a range of binary health-related questions. Our evaluation considers various context and prompt conditions, with the objective of determining the impact of these factors on the quality of the responses. Additionally, we explore the effect of in-context examples in the performance of top models. To further validate the obtained results, we also conduct contamination experiments that estimate the possibility that the models have ingested the benchmarks during their massive training process. Finally, we also analyse the main classes of errors made by these models when prompted with health questions. Our findings contribute to understanding the capabilities and limitations of LLMs for health information seeking.

This study investigates the impact of advanced computational methodologies on brain tumor segmentation in medical imaging, addressing challenges like interobserver variability and biases. The DeepLabV3plus model with ResNet50 integration is rigorously examined and augmented by diverse image enhancement techniques. The hybrid CLAHE-HE approach achieves exceptional efficacy with an accuracy of 0.9993, a Dice coefficient of 0.9690, and a Jaccard index of 0.9404. Comparative analyses against established models, including SA-GA, Edge U-Net, LinkNet, MAG-Net, SegNet, and Multi-class CNN, consistently demonstrate the proposed method’s robustness. The study underscores the critical need for continuous research and development to tackle inherent challenges in brain tumor segmentation, ensuring insights translate into practical applications for optimized patient care. These findings offer substantial value to the medical imaging community, emphasizing the indispensability of advancements in brain tumor segmentation methodologies. The study outlines a path for future exploration, endorsing ensemble models like U-Net, ResNet-U-Net, VGG-U-Net, and others to propel the field toward unprecedented frontiers in brain tumor segmentation research.

This paper approaches a new model for arrhythmia diagnosis based on short-duration electrocardiogram (ECG) heartbeats. To detect 8 arrhythmia classes efficiently, we design a Deep Learning model based on the Focal modulation layer. Moreover, we develop a distance variation of the SMOTE technique to address the problem of data imbalance. The classification algorithm includes a block of Residual Network for feature extraction and an LSTM network with a Focal block for the final class prediction. The approach is based on the analysis of variable-length heartbeats from leads MLII and V5, extracted from 48 records of the MIT-BIH Arrhythmia Database. The methodology’s novelty consists of using the Focal layer for ECG classification and data augmentation with DTW distance (Dynamic Time Warping) using the SMOTE technique.The approach offers real-time classification and is simple since it combines feature extraction, selection, and classification in one stage. Using data augmentation with SMOTE variant and Focal-based Deep learning architecture to identify 8 types of heartbeats, the method achieved an impressive overall accuracy, F1-score, precision, and recall of 98.61%, 94.08%, 94.53%, and 93.68% respectively. Additionally, the classification time per sample was only 0.002 s. Therefore, the suggested approach can serve as an additional tool to aid clinicians in ensuring rapid and real-time diagnosis for all patients with no exclusivities.

Medical data often presents as a time series, reflecting the disease's progression. This can be captured through longitudinal health records or hospital treatment notes, encompassing diagnoses, health states, medications, and procedures. Understanding disease evolution is critical for effective treatment. Graph embedding of such data is advantageous, as it inherently captures entity relationships, offering significant utility in medicine. Hence, this study aims to develop a graph representation of Electronic Health Records (EHRs) and combine it with a method for predictive analysis of COVID-19 using network-based embedding. Evaluation of Graph Neural Networks (GNNs) against Recurrent Neural Networks (RNNs) reveals superior performance of GNNs, underscoring their potential in medical data analysis and forecasting.

Survival analysis focuses on the prediction of failure time and serves as an important prognostic tool, not solely confined to medicine but also across diverse fields. Machine learning methods, especially decision trees, are increasingly replacing traditional statistical methods which are based on assumptions that are often difficult to meet. The paper presents a new global method for inducing survival trees containing Kaplan–Mayer estimators in leaves. Using a specialized evolutionary algorithm, the method searches for oblique trees in which multivariate tests in internal nodes divide the feature space using hyperplanes. Specific variants of mutation and crossover operators have been developed, making evolution effective and efficient. The fitness function is based on the integrated Brier score and prevents overfitting taking into account the size of the tree. A preliminary experimental verification and comparison with classical univariate trees was carried out on real medical datasets. The evaluation results are promising.

Venous thromboembolism (VTE) poses a significant risk to patients undergoing cancer treatment, particularly in the context of advanced and metastatic disease. In the realm of neuro-oncology, the incidence of VTE varies depending on tumor location and stage, with certain primary and secondary brain tumors exhibiting a higher propensity for thrombotic events. In this study, we employ advanced machine learning techniques, specifically XGBoost, to develop identifying models for predictors searching associated with VTE risk in patients with gliomas. By comparing the diagnosis testing accuracy of our XGBoost models with traditional logistic regression approaches, we aim to enhance our understanding of VTE prediction in this population. Our findings contribute to the growing body of literature on thrombosis risk assessment in cancer patients and may inform the development of personalized prevention and treatment strategies to mitigate the burden of VTE in individuals with gliomas at the hospital term.