Bioinformatics and Biomedical Engineering

In cancer treatment with radiation, the objective is to kill tumor cells by damaging their DNA. However, the DNA strand breaks are not yet understood by which types and energy of electron interactions they are caused. Most of Monte Carlo simulations reported in the literature consider an energy deposition in water-equivalent to cause DNA strand breaks. In recent years, DNA atomistic models were introduced but still the simulations consider energy deposition in volumes of DNA material. In the present work, we report advances in the understanding of DNA single and double strand breaks by low energy electrons, and the results of the simulations were compared to experimental data. We simulated a model of atomistic B-DNA, forming 1122 base pairs twisted and forming a length of 30 nm. Each atom has been represented by a sphere whose radius is equal to the radius of van der Waals. We repeatedly simulated 10 million electrons for each of the energies from 4 eV to 500 eV and counted each interaction type with its position x,y,z in the volume of DNA. Based on the number and types of interactions at the atomic level, the number of DNA single and double strand breaks were calculated. In addition, with our simulation, it is straightforward to discriminate the strand and base breaks as a function of radiation interaction type and energy. In conclusion, the knowledge of DNA damage at the atomic level helps, for example, to design internal therapeutic agents of cancer treatment.

Metabolite analysis reveals new challenges in human health care. This human health care connects to the immune system and presents opportunities for the prevention and detection of early hidden disease symptoms. Predicting the concentration of immune metabolites and confirming relationships between concentrations of individual metabolites have the potential to create breakthroughs in diagnostic techniques. This early detection of serious diseases plays a major role in overall recovery. Moreover, metabolite analysis linked to biomedical applications could provide an ideal tool for preventive healthcare and the pharmaceutical industry.This study presents the linear prediction of selected metabolites involved in the immune system. The evaluation relied on accurate linear prediction modeling and subsequent comparison. This is the first step toward determining the relationship of metabolites and immune system using computational biomedical analysis.

Numerical simulation is a tool used in multiple scientific domains. There is a wide range of simulations where we model a flow of fluid in a specific geometry, for example in simulations of blood flow in microfluidic channels. In such cases, a complex shape of channels has to be defined by describing its boundaries and rigid obstacles. The purpose of this study is develop a method of defining boundaries and obstacle objects with complex and non-trivial shapes in such numerical simulations. The obstacle or a boundary needs to be described only by a cloud of points defining its surface. Based on this point cloud a distance function determining the position and the shape of the obstacle is defined in the whole simulation domain. This general method is presented on a concrete examples involving several simulations performed within a simulation package ESPResSo. The new method of obstacle creation gives excellent results in terms of the accuracy and simulation time consumption.

Childhood obesity is a multifactorial disease influencing the development of a range of metabolic disorders, where adipose tissue has been proved to be fundamental. The adipose tissue can be distributed throughout the body as visceral adipose tissue (VAT) and subcutaneous adipose tissue (SAT), and there are considerable anatomical differences between both adipose tissues in the body. Importantly, VAT is associated with low-grade systemic inflammation and insulin resistance, which are key factors underlying metabolic alterations associated with childhood obesity [1]. This study aimed to identify the molecular signatures underlying obesity and overweight in children, differentiating between shared and individual signatures in VAT and SAT. Both tissue samples were collected from 18 children (11 girls) aged 0.54 to 16.63 years and hospitalized for abdominal surgery, of which 6 children (2 girls) had overweight or obesity. RNAseq analysis was performed to identify gene expression patterns associated with obesity and overweight in each tissue. The software tools used in the RNAseq data analysis were FastQC, to perform sequencing quality checks; HISAT, to map reads to the human genome; featureCounts to quantify raw counts; and DESeq2 to differential gene expression analysis. In VAT there were 759 genes showing statistically significant differential expression between groups (nominal p-value < 0.05), from which 48 passed an FDR threshold of 0.05. VAT’s differential expression results may be observed in Fig. 1–2. In SAT there were 945 genes showing statistically significant differential expression, from which 28 passed the FDR threshold. SAT’s differential expression results are shown in Fig. 3–4. We were specially interested in the identification of shared genes associated with overweight and obesity in both tissues, for which we performed a gene ontology analysis of all differently expressed genes, using Gene Ontology and Kyoto Encyclopedia of Genes and Genomes databases. Among significantly associated genes, there were 126 common genes, as it is shown in Fig. 5. The reliability of our results was assessed by matching our list of significant genes with a list of genes whose influence on obesity has been previously described in the literature (e.g., LEP and TNMD) [2, 3]. Additionally, our research identified new molecular targets, highlighting the results of VAT (i.e., XIST, PRKY and TTTY10). In conclusion, our approach identified independent and shared gene expression patterns in VAT and SAT associated with overweight and obesity in children. Understanding the molecular architecture of obesity with approaches like these is crucial for the identification of powerful molecular targets and developing of effective precision medicine therapies.

Astrocytes are a particular type of glial cells observed throughout the gray matter in the brain. In a healthy brain, they help to defend evolutionarily conserved astrogliosis programs and maintain neuronal metabolism. On the other hand, in the Alzheimer’s disease (AD) affected brain, they release neurotoxins because of the adopting behaviours of different functions depending on the disease progression. Along with astrocytes, amyloid-beta (A $$\beta $$ β ) and tau proteins ( $$\tau $$ τ P) play a prominent role in AD. In this paper, we have developed a model and have studied the dual action of astrocytes with A $$\beta $$ β , $$\tau $$ τ P, and their toxic forms in the brain connectome. Initial conditions-dependent solutions of the model demonstrate that the treatment depends on AD’s status at the first diagnosis time. With an increase in the clearance rate of toxic A $$\beta $$ β by the astrocytes, the model predicts a cure possibility from AD. Furthermore, the network model with non-uniform parameter values in different regions, developed here, provides a better insight into the distributions of the concentrations in the brain connectome.

Working memory (WM) has been intensively used to enable the temporary storing of information for processing purposes, playing an important role in the execution of various cognitive tasks. Recent studies have shown that information in WM is not only maintained through persistent recurrent activity but also can be stored in activity-silent states such as in short-term synaptic plasticity (STSP). Motivated by important applications of the STSP mechanisms in WM, the main focus of the present work is on the analysis of the effects of random inputs on a leaky integrate-and-fire (LIF) synaptic conductance neuron under STSP. Furthermore, the irregularity of spike trains can carry the information about previous stimulation in a neuron. A LIF conductance neuron with multiple inputs and coefficient of variation (CV) of the inter-spike-interval (ISI) can bring an output decoded neuron. Our numerical results show that an increase in the standard deviations in the random input current and the random refractory period can lead to an increased irregularity of spike trains of the output neuron.

The aim of this pilot study is to know the thermal impact that manual therapy has on the body temperature of the lower back and abdominal region in subjects with low back pain. It was conducted on ten patients, five of them diagnosed with low back pain and five without it. The intervention protocol was based on Richelli’s instrumentalized manual therapy, it was carried out by a certified physical therapist, one session per week, for two weeks. Infrared Thermography was used to show the effects of this study. The regions of interest considered for the thermal analysis were the rectus abdominis, the obliques, the quadratus lumbar and the lumbar spine. The temperature difference between post and pre-intervention in patients with low back pain was higher in the first week, maintaining differences greater than 0.78, while in the second week the maximum difference was 0.52 in the lumbar area. As a result, the instrumentalized manual intervention technique tends to stabilize the temperature of the muscles, managing to favor the subjects with low back pain because this specific group achieved greater temperature stability in the intervened muscle groups. Infrared Thermography proved to be a useful tool for monitoring physiotherapeutic treatments.

The main aim of this study was to compare bone health parameters (bone mineral content [BMC], bone mineral density [BMD], geometric indices of femoral neck [FN] strength (cross-sectional area [CSA], cross-sectional moment of inertia [CSMI], section modulus [Z], buckling ratio [BR] and strength index [SI]) and composite indices of FN strength (compression strength index [CSI], bending strength index [BSI], and impact strength index [ISI])) in young adult inactive women (n = 18) and young adult female handball players (n = 20). The participants of the study were 38 young adult women; their ages ranged from 20 to 32 years. Dual-energy X-ray absorptiometry (DXA) was used to evaluate body composition, BMC, BMD and geometric indices of FN strength. BMD measurements were completed for the whole body (WB), the lumbar spine (L1–L4), the total hip (TH) and the femoral neck (FN). Composite indices of femoral neck (FN) strength (CSI, BSI, and ISI) were calculated. Validated tests were used to evaluate maximal bench-press strength, maximal half-squat strength, maximal deadlift strength, vertical jump, horizontal jump, sprinting performance and maximal aerobic velocity. Validated questionnaires were used to evaluate sleep quality, daily protein and calcium intakes and physical activity level. Lean mass, WB BMC, WB BMD, L1–L4 BMD, TH BMD, FN BMD, CSMI, Z, physical activity level, daily protein intake, daily calcium intake, one-repetition maximum (1-RM) bench press, 1-RM half-squat and 1-RM deadlift were significantly higher in female handball players compared to inactive women. In conclusion, the present study suggests that handball practice is associated with better bone health parameters in young adult women.

Finite element models (FEM) have been a breakthrough in the field of medicine for a wide variety of applications. They have been used, for example, for predicting the behaviour of many biological structures, as well as to foresee the possible outcomes of some types of operations. One of the basic problems when modelling biological structures is finding a way to determine the initial geometric parameters with a sufficient degree of precision, so that the results are representative. In the case of computational models used for the study of corneal biomechanics, the knowledge of initial conditions defined in a finite element model become critical, since they represent the in-vivo state of the biological structure by means of a computer simulation. There is a lack of consensus among the investigations carried out to date regarding whether the initial status in the FEM models should be considered or not. In this research work, two approaches that aim to determine the geometry of the in-vivo state of the cornea with mild keratoconus have been compared: the so-called stress-free geometry on the one side, and the initial tension state on the other side. The results obtained allow comparisons between them, and validate both approaches when they are used to obtain corneal geometry in initial stress-free conditions for a FEM model when incipient keratoconus.

Bilateral symmetry in the human body is a necessary feature for our body to function more efficiently. Apparently, it could be thought that the ocular structure presents a bilateral symmetrical structure, however this does not occur in all cases, especially in those that are pathological. In many cases, symmetry is essential for the achievement of certain optical tasks. This study presents a method for evaluating bilateral symmetry (direct vs mirror) from the differences between the Cartesian coordinates (OD vs OS) in healthy corneas using customized patient-specific models and unprocessed data from corneal tomographs. The results obtained in the central and paracentral regions of the corneas evaluated show that the bilateral mirror symmetry presents differences below 10 microns and the direct bilateral symmetry reaches differences of the order of 30 microns at spatial coordinates level. This study shows that the level of asymmetry in healthy corneas can be evaluated by direct or mirror symmetry, and that the presence of asymmetric anomalies can be useful as a clinical diagnostic tool.

Small animals stroke models have widely been used to study the mechanisms of ischemic brain damage in controllable experimental settings. The evaluation of stroke lesions mainly relies on visual inspection of tissue samples collected after brain sectioning, slice staining and scanning, a procedure that is highly subjective and prone to human error. In this study we developed a machine-learning based methodology for automatic segmentation of lesions in mouse brain tissue samples, stained with Triphenyltetrazolium chloride (2% TTC). Our approach relies on the creation of a statistical mouse brain atlas of healthy TTC slices that was lacking in the literature. For this purpose we applied tissue clustering and Markov Random Fields (MRF) for brain tissue detection followed by deformable image registration for spatial normalization. The obtained statistical atlas is then exploited by outlier detection techniques and Random Forest classification to extract lesion probability maps in new slices. The good agreement between our segmentation results and expert-based lesion delineation on 12 mouse brains highlights the potential of the proposed approach to automate stroke volumetry analysis, thereby contributing to increased translational capacity of experimental stroke.

To present time, the world science has accumulated a sufficient amount of information on the quantitative changes in the rate of glucose utilization according to positron emission tomography (PET) with 18-fluorodeoxyglucose (FDG) in Parkinson’s disease (PD) with dementia. It was found that in the early stages of dementia with PD, there is a dysfunction of the frontal lobes, while PET examination of patients with PD with dementia shows the quantitative changes in the rate of glucose utilization. On this basis, it seems relevant to compare the functional state of the brain structures and the results of neuropsychological studies in patients with PD with varying severity of cognitive impairment. The present research was aimed to investigate the relationship between early cognitive violations according to neuropsychological research and the rate of glucose metabolism in different brain areas during PET scanning in patients with Parkinson’s disease. Neuropsychological testing consisted of clinical interviews, observation, questioning, Mini-Mental State Exam (MMSE) and frontal assessment battery (FAB). According to the research outcomes, it was found that with initial cognitive impairment, determined by FAB, a quantitative change in the rate of glucose utilization is observed, similar to the pattern found in patients with cognitive disorders in PD. Four factors were established: factor 1 – dorsal system of attention (voluntary attention), factor 2 – ventral system of attention (involuntary attention), factor 3 – system of the state of operational rest, factor 4 – visual projective zone. The precentral cortex of the frontal areas and the upper half of the parietal zones of big brain are constitute the 1st factor. The anterior third of the convexital part of the frontal areas and the lower half of the parietal zones are the 2nd factor. The 3rd factor includes the 23, 36, 29 and 30 cytoarchitectonic Brodmann fields (the posterior cingulate gyrus). Primary visual area (17th Brodmann’s field) is the 4th factor.

Electrical impedance myography (EIM) is a non-invasive measurement technique capable of determining physiological and morphological changes in skeletal muscle tissue. The method consists of applying an alternating current on the skin through two electrodes to determine the bioimpedance of the medium. EIM has contributed to the study of degenerative diseases that changes the electrical properties of muscle tissue during its evolution of the disease. The objective of this work is to develop a portable and low-cost impedance measurement instrument to assist in activities related to the study and diagnosis of muscle diseases. We present a modular hardware architecture that allows the impedance measurement of multiple muscles at the same time, but in the present work we test the performance of the prototype with a single impedance measurement channel implemented. The tests were performed with known theoretical impedance calculated to evaluate the accuracy of the impedance meter, as well as other performance tests of the system.

Data-driven science and its consequences in the extensive field of artificial intelligence and especially in machine learning have the potential to drive important changes in medicine. Therewith, new instances including new incoming mixed features are unceasingly emerged at high rate as data streams. Hence, big data promises immense advantages for medical and healthcare research. Aiming to analyse these medical data streams in the lastly mentioned fields, this paper provides a better insight into an advanced incremental attribute and object learning k-prototypes algorithm for healthcare and medical applications. Experiments performed on various real mixed healthcare data sets show that the proposed real-time healthcare/medical application is efficient and may cover different medical case studies such as patient monitoring, disease control, and clinical support systems for better prediction of diseases. The measured evaluation criteria accentuate the efficiency of the proposed method compared to the conventional k-prototypes method.

Background: The goal of this work was to measure artery inflammation in aged volunteers with atherosclerosis using computed tomography (CT) and positron emission tomography (PET) with 18F-FDG. The artery plaques are composed of lipid rich fibrous tissue and foamy macrophages and are the most vulnerable for detachment. Such plaques can be differentiated by their density with CT imaging. Methods: A healthy artery (NAR) was considered with no plaque on a CT images and with density between 51 and 130 Hounsfield Units (HU). A non-calcified plaque (NCP) and a calcified plaque (CP) were respectively identified as having a density ≤ 50 HU and >130 HU. In the calcified arteries, the calcification area divided by the artery area (RCA) and the calcification score (ACS) were classified with Hierarchical K-means algorithm into 4 clusters and were correlated with the metabolic rate of 18F-FDG (MRG). Results: we found MRG statistically higher in NCP in comparison to NAR and CP in subjects without medication (P < 0.05). In subjects under-medication, NCP values were found the lowest. MRG of NCP in non-medication subjects was statistically significantly different from CP with small area but not from CP with large areas (P = 0.40). In under-medication subjects, no statistical differences were found between NCP and CP independently of plaque area and density. Conclusion: Since the low-density plaque was reported as the vulnerable plaque, based on the present work, this latter can be simply identified on CT images with intensity between 30 HU and 50 HU.

Statistically revealed inner structuredness of bacterial vs. chloroplast phototsynthetic genes is studied. To do it, we analysed the cyanobacterial genes responsible for a light consumption. A sounding difference in the spatial pattern specific for chloroplast photosystem genes, and bacterial photosystem genes is found. Thus, the bacterial genes yield another type of symmetry of the distribution of the genes converted into triplet frequency dictionaries in 63-dimensional Euclidean space of triplets.

We used unsupervised nonlinear clustering to reveal the interplay between structure of nucleotide sequences and the taxonomy of their bearers. Triplet frequency composition is referred to a structure, and taxonomy is determined through standard morphology and physiology of bacteria. Soft $$16\times 16$$ 16 × 16 elastic map has been used for clustering. Some preliminary results are presented here approving the high efficiency of such approach to phylogeny analysis. Further applications to medicine are discussed.

Cancer remains the most common worldwide problem with the highest impact on global health. It is the second leading cause of death, due to the lack of early diagnosis and high recurrence rate after conventional therapies. Although every year several new therapeutic approaches are proposed the urgent need for more effective therapeutic strategies to improve the survival rate and life expectancy of cancer patients rapidly grows.A recent promising anticancer strategy is based on multinuclear heterocycles as widely investigated bioactive molecules, considered important synthetic targets for the development of novel therapeutic agents. Many nitrogen heterocycles are known for a long time as natural alkaloids, known to possess the broad and diverse biological activity and medicinal applicability. Nowadays however novel multinuclear drug-like heterocyclic structures are generated by methods of artificial intelligence. Novel approaches are required as more expeditious ways of studying their biological activity, capable of more than explaining their activity, and even prognosticating it.This study highlights our and other authors’ recent results on the biological activity of multinuclear heterocyclic molecules on cancer cells, explicitly based on their capacity to bind to G-quadruplexes. It further stresses the need for novel G-quadruplex binding compounds, with elucidated biochemical mechanisms of action for biomedical applications, namely in anticancer therapies.

This paper describes FIMED 2.0, a Service for Flexible Management and Analysis of Heterogeneous Clinical Data. This software tool allows flexible clinical data management from multiple trials, which can help to improve the quality of clinical data and ease in clinical trials. The proposed service has been developed on top of a NoSQL Database (MongoDB), which allows for collecting and integrating clinical data in dynamic and incremental schemes based on their needs and clinical research requirements. Building upon our experiences with Flexible Management of Biomedical Data (FIMED), we have developed this new version of the tool aiming not only at replicating the former one but also including further gene regulatory network analysis and data visualization oriented to annotate gene functionality and identify hub genes. This version allows the practitioner to use four different network construction methods such as data assimilation, linear interpolation, tree-based ensemble or Gradient Boosting Machine regression. You may find a free version of this tool on the web at https://khaos.uma.es/fimedV2 . A demo user account has been created to provide user demonstration, “iwbbio”, using the password “demo”. A real-world use case for a clinical assay in Melanoma disease is also included in this demo, which has been indeed anonymized.

There are several situations where it is necessary to acquire analog signals, like sensors outputs and bioelectrical signals, with different amplitudes and frequency ranges. The FPAA (Field Programmable Analog Array) is a semiconductor device that allows the creation of several analog circuits. The Arduino is a platform for rapid development with microcontrollers, well known and widely used to build experimental and commercial equipment. Given the above, the objective of this work was to create a hardware board (shield) and a software library to use an FPAA in conjunction with Arduino boards. In one test, we implemented a band-pass filter and obtained between the projected and measured frequency response an average error of 0.027 dB (SD = 0.163 dB). The maximum error was 0.265 dB. In another test, we implemented a circuit to capture the ECG signal. The results of the ECG test were satisfactory. This research introduces a significant contribution to bioelectrical signal acquisition since similar works do not exist.

This review paper focuses on analyzing research work related to the utilization of smartwatches in health informatics. In recent years, we have seen an ascent in life expectancy due to considerable innovations in the healthcare industry. Sicknesses identified with the cardiovascular framework, eye, respiratory framework, skin, and emotional well-being are inescapable around the world. Most of these sicknesses can be kept away from or potentially appropriately oversaw through consistent examining. To empower ceaseless well-being checking to serve developing medical care needs, moderate, non-intrusive, and simple to-utilize medical services arrangements are basic. The increasing use of wearables watches coupled with health monitoring sensors makes it an essential tech for a continuous and remote health examination. In this paper, we present a comprehensive review of different research work on the utilization of smartwatches to deal with various diseases. For this, we have screened 370 research publications related to smartwatches in health informatics and selected 20 journals for the review that matched our selection criteria. Finally, we discussed future research perspectives and concerns regarding smartwatch-enabled healthcare architecture.

With the exponentially growing amount of data collected by wearable medical devices, there is an expansion of opportunities to exploit machine learning methods in monitoring patient health and predicting events that may need medical attention. Wearable ECG devices provide a more comfortable alternative compared to the usual monitoring devices used in clinical settings, at the cost of inferior information density and signal quality. Notwithstanding, recent studies suggest that machine learning methods are able to work with wearable data, and for specific tasks, even medical grade devices are now available (e.g. detection of cardiac arrhythmias). This paper focuses on improving the quality of machine learning data, as part of an ongoing research of automating pipelines that allow for building robust models based on wearable ECG signals. While our efforts in general also consider the mainstream machine learning task of heartbeat classification, this work instead (exploiting the long-term ECG data) contributes to the development of models to learn features much longer than a single heartbeat. For the extraction of such features, R-peak detectors are considered and evaluated with respect to sensitivity and robustness to noise. Noise stress tests show that a combination of global filters can improve the performance of detectors, efficiently dealing with typical noises almost always present in wearable ECG signals. Also, local noise detection models are demonstrated to be promising methods to handle heavy noises that can not be resolved by the global filters.

The presented paper is focused on possibility to evaluate echogenicity grade on ultrasound B-images of the thyroid gland tumor. We use our developed software tool which has been originally designed for analysis of ultrasound B-images in neurology. Currently, the goal of this research is to decide if this software is also useful for different ultrasound B-images, in this case in endocrinology. To evaluate it, the Echo-Index parameter is used. The core principle of the algorithm is based on computing area and computed echogenicity index. The paper has two main parts. The first one is focused on general assessment of the reproducibility of computed Echo-Index between two independent, non-experienced observers. The second part is concentrated on changes of the echogenicity index (Echo-Index) in the case of malignant tumor and in the case of benign tumor. The general reproducibility is almost perfect, level of agreement based on correlation coefficient is higher than 0.98 and the absolute average error does not exceeds 15% between two observers. Subsequent analysis between malignant and benign tumor images evinces the Echo-Index is different.

Even though cancer is one of the most common diseases in the 21st century, early detection tests are still expensive and invasive. In this work, a initial stool monitoring prototype for the early detection of this disease is proposed. The emerging and growing concept of Internet of Things has been considered for the implementation of this prototype. So, MOX (Metal OXide, metal-oxide-semiconductor type) sensors to detect volatile organic compounds (VOCs) and a thermal camera are integrated into different development boards, which send the collected data of stool to monitoring it by means of an IoT platform. With the result of this initial prototype, a proof of concept has been obtained for testing with cancer experts.

Background: Developmental coordination disorder (DCD) is a neurodevelopmental disorder encountered in about 6% of children at school age. DCD mostly affects motor task automatization and it persists in adulthood. Several brain structures were supposed to be involved in DCD pathophysiology. Quantitative imaging techniques have the potential to investigate these connected brain regions associated with motor tasks. Methods: In the present work, we studied with 18F-fluorodeoxyglucose (18F-FDG) and positron emission tomography imaging (PET) brain metabolism in subjects with DCD versus control in resting state and during repetitive and standardized finger movements of the left non-dominant hand. We analyzed 42 brain structures in the right and left hemispheres and the data were statistically assessed by the Ward clustering approach to detect the activated/non-activated/deactivated brain regions. Results: The images obtained with PET clearly showed different uptake of 18F-FDG in subjects with DCD with respect to control subjects. The statistics showed less brain regions activated and more deactivated in subjects with DCD than in control. Among other differences, the right thalamus was activated in DCD subjects as both caudate nuclei were deactivated for a possible compensation for basal ganglia dysfunction or deficit. Conclusions: This first PET study of DCD found significant thalamus activation as previous studies on finger movement tasks comparing PET and fMRI in normal subjects. The Ward clustering in DCD images allowed to identify activated/non-activated/deactivated brain structures in subjects with DCD versus control.

In the last few years, eXplainable Artificial Intelligence (XAI) has been attracting attention in data analytics, as it shows great potential in interpreting the results of complex machine learning models in the application of medical problems. The nutshell is that the outcome of the machine learning-based applications should be understood by end users, specially in medical data context where decisions have to be carefully taken. As such, many efforts have been carried out to explain the outcome of a deep learning complex model in processes where image recognition and classification are involved, as in the case of Melanoma cancer. This paper represents a first attempt (to the best of our knowledge) to experimentally and technically investigate the explainability of modern XAI methods Local Interpretable Model-Agnostic Explanations (LIME) and Shapley Additive exPlanations (SHAP), in terms of reproducibility of results and execution time on a Melanoma image classification data set. This paper shows that XAI methods provide advantages on model result interpretation in Melanoma image classification. Concretely, LIME performs better than SHAP gradient explainer in terms of reproducibility and execution time.

In this work, we propose the use of thermal vision sensors to estimate the frontal body landmarks of an inhabitant. The use of thermal sensors is being promoted to collect human patterns while protecting inhabitants’ privacy in smart environments. On the other hand, deep learning approaches have provided encouraging results in estimating body, hand and facial landmarks. Here, we present a residual neural network which produces body landmarks from images collected by a low cost thermal sensor. In order to solve the problems of capturing and labeling data, which hinder learning in deep learning models, we propose an auto-labeling approach with dual visible-spectrum and thermal cameras, including the recognition of keypoints by the OpenPose model. A case study developed with four inhabitants in different poses shows encouraging results.

In current ophthalmology, images of the vascular system in the human retina are used as exploratory proxies for pathologies affecting different organs. In this brief paper, we use multi-modal retinal images for assisting diagnostic decision making in diabetes mellitus and diabetic retinopathy. We report the use of matrix factorization-based source extraction techniques to pre-process the images as a data quality control step prior to their classification. Through this quality control, we unveil some relevant sources of bias in the data. After correcting for them, promising pathology classification results are still obtained, which merit further exploration.

Radiomics is a systematic approach to characterize objects in terms of their radiological appearance. We used radiomic features of 5027 objects of 6 classes and trained a binary classifier with 79% accuracy. Features were obtained by using our novel preprocessing pipeline for object segmentation from the lung tissue in a low-dose Computed Tomography (LDCT) imaging technique. Our results show that radiomic features prove effective in distinguishing between suspicious and benign objects located in the lung tissue. Our data shows that there is vast space for improvement from both model- as well as a data-centric approach to developing Computer-aided detection (CAD) systems based on radiomics for early lung cancer detection. We show our results in the paper.

Segmentation of brain images generated by magnetic resonance imaging (MRI) is an important part of clinical medicine as it enables three-dimensional reconstruction and downstream analysis of normal and pathological regions. Segmenting white matter (WM), grey matter (GM), and cerebrospinal fluid (CSF) automatically are challenging tasks. In this paper, a clustering-based segmentation of MR images is performed using a modified quantum-inspired firefly algorithm with mutation operation. In the proposed method, a mutation operation based on the X-gate has overcome the restriction on initial centroids trapped in local optima. The objective function is chosen to be the minimum intra-cluster distance. The suggested approach has been tested on several sections of human brain images with differing cluster numbers. Correlation, SSIM, entropy, and PSNR have been used to evaluate the outputs of the method. The evaluation metrics indicate that the proposed clustering-based algorithm successfully segmented the MR images.

Sleep Disorders have received much attention in recent years, as they are related to the risk and pathogenesis of neurodegenerative diseases. Notably, REM Sleep Behaviour Disorder (RBD) is considered an early symptom of $$\alpha $$ α -synucleinopathies, with a conversion rate to Parkinson’s Disease (PD) up to 90%. Recent studies also highlighted the role of disturbed Non-REM Slow Wave Sleep (SWS) in neurodegenerative diseases pathogenesis and its link to cognitive outcomes in PD and Dementia. However, the diagnosis of sleep disorders is a long and cumbersome process. This study proposes a method for automatically detecting RBD from single-channel EEG data, by analysing segments recorded during both REM sleep and SWS. This paper inspects the underlying microstructure of the two stages and includes a comparison of their performance to discuss their potential as markers for RBD. Machine Learning models were employed in the binary classification between healthy and RBD subjects, with an 86% averaged accuracy on a 5-fold cross-validation when considering both stages. Besides, SWS features alone proved promising in detecting RBD, scoring a 91% sensitivity (RBD class). These findings suggest the applicability of an EEG-based, low-cost, automatic detection of RBD, leading to potential use in the early diagnosis of neurodegeneration, thus allowing for disease-modifying interventions.

Alzheimer disease (AD) is the most common form of senile brain disorder. AD is not reversible, but its neuropathology can be detected several years before severe clinical manifestations. AD diagnosis is carried out relying on several clinical data, such as MRI structural and functional data, PET and DTI imaging, neuropsychological tests’ scores, genetic data, and others. Approaches that use complementary information and heterogeneous sources of data might have a decisive impact on the ability to early identify and consequently treat those subjects with a higher probability of conversion. We propose an on-going work on a Deep Learning framework that integrates different sources of data such as imaging data, clinical data, neuropsychological tests’ scores, and the temporal information related to the last medical evaluation of the subject, with the aim of estimating the probability of conversion from mild-cognitive-impairment (MCI) to AD or from a stable clinical profile to MCI in a period of time that varies from 6 months to 18 months. The possibility of predicting disease conversion is an open problem in this field, and wants to answer to a specific need of clinicians. The ADNI public data-set represents the reference data-set: an extensive and detailed analysis of ADNI has been performed to assess the sample size available for the training and testing of the network, that is now under construction, and the first results will be soon available. The network will also be tested with clinical data of the Fondazione Santa Lucia, Rome (Italy), and results will be discussed with the neurologists, neuropsychologists, and physics that are actively working with us.

The precision and personalized medicine is declared as the next revolutionary paradigm in the current health outlook. With this assumption, many challenges must be faced to achieve that paradigm shift. One of these important challenges is the creation and development of tools with the capability of exploiting biological data to infer or extract new and relevant knowledge. In this sense, these tools must fulfill a set of requirements such as scalability, security and a user-friendly design. That is the way to change the users scope from technicians to all type of researchers, physicians and other non-technical users. Along this article, a review of several gene expression analysis tools has been addressed, with the aim of studying their pros and cons. Then, two different implementations are proposed taking into account the current state of KnowSeq R/Bioc package, with the purpose of showing different use cases to migrate one concrete tool to a web application.

The coronavirus disease (COVID-19) pandemic has challenged multiple aspects of our lives. Social distancing among other preventive measures for reducing the contagion probability have supposed a significant challenge for many establishments. Restaurants, schools, conferences are establishments founded by the congregation of participants, distributed in tables or chairs over a certain scenario. These enterprises now face an optimization problem in their daily routine, where they seek to maximize the interpersonal distance while also allocating the maximum number of assistants. The optimization of these distribution paradigms, such as the CLP (Chair Location Problem), has been defined as NP-Hard, therefore, the use of metaheuristic techniques, such as Genetic Algorithms is recommended for obtaining an optimal solution within a polynomial time. In this paper, a GA is proposed for solving the CLP, attaining an optimal solution that maximizes the interpersonal distance among assistants while also guaranteeing a minimum distance separation for reducing the contagion probability. Results of the proposed methodology and multiple fitness evaluation strategies prove its viability for attaining a valid distribution for these establishments, thus satisfying the main objectives of this research.

Established text mining approaches can be used to identify miRNAs mentioned in published papers and preprints. Here, we apply such a targeted approach to the LitCovid literature collection in order to find viral miRNAs published in connection to SARS-CoV-2. As LitCovid aims at being a comprehensive collection of literature on new findings on SARS-CoV-2 and the COVID-19 pandemic, it is perfectly suited for our goal of finding all reported SARS-CoV-2 miRNAs. The identified miRNAs provide an up-to-date and quite comprehensive collection of potential viral miRNAs, which is a useful resource for further research to fight the current pandemic.We identified 564 putative SARS-CoV-2 miRNAs together with the respective evidences, i.e. the original publications, and collect them for critical review. The text mining method and the corresponding synonym list are optimized for finding viral miRNAs and the results are manually curated. Since not all miRNAs were experimentally verified, the collection might contain false positives, but it is highly sensitive. Moreover, the text mining approach and resulting collection of miRNA candidates can be useful resources for further SARS-CoV-2 research and for experimental validation.

One of the most important situations in recent years has been originated by the 2019 Coronavirus disease (COVID-19). Nowadays this disease continues to cause a large number of deaths and remains one of the main diseases in the world. In this disease is very important the early detection to avoid the spread, as well as to monitor the progress of the disease in patients, and techniques of artificial intelligence (AI) is very useful for this. This is where this work comes from, trying to contribute in the study to detect infected patients. Drawing inspiration from previous work, we studied the use of deep learning models to detect COVID-19 and classify the patients with this disease. The work was divided into three phases to detect, evaluate the percentage of infection and classify patients of COVID-19. The initial stage use CNN Densenet-161 models pre-trained to detects the COVID-19 using multi-class X-Ray images (COVID-19 vs. No-Findings vs. Pneumonia), obtaining 88.00% in accuracy, 91.3% in precision, 87.33% in recall, and 89.00% in F1-score. The next stage also use CNN Densenet-161 models pre-trained to evidenced the percentage of infection COVID-19 in the different CT-scans slices belonging to a patient, obtaining in the evaluation metrics a result of 0.95 in PC, 5.14 in MAE and 8.47 in RMSE. The last stage creates a database of histograms of different patients using their lung infections and classifies them into different degrees of severity using K-Means unsupervised learning algorithms with PCA.

The aim of this study was to check the role of information sources, trust in information sources and COVID-19 related conspiracy in the compliance with COVID-19 related measures at the beginning of the pandemics and at the moment in which study was conducted, on Serbian and German sample. Previous studies have found that information and conspiracy theories are important variables in predicting compliance with COVID-19 related measures, and that Serbian and German culture differ in these behaviors. Instruments used were questions regarding Information sources, questions regarding Trust in information sources (both separated into Formal and Informal), question measuring belief that COVID-19 was created in the laboratory, and two question measuring self-assessed compliance with COVID-19 related measures. Data was analyzed with moderated path analysis through structural equation modeling. The results of this study show that Trust in formal sources negatively predicted belief that COVID-19 was created in the laboratory and positively compliance with COVID-19 related measures, on both samples. On Serbian sample Informal sources also negatively predicted belief that COVID-19 was created in the laboratory, while this belief predicted (negatively) compliance with COVID-19 measures at the beginning of pandemics. We can conclude that there are some cultural differences, and that COVID-19 related conspiracy is more important on Serbian than on the German sample.