Search

Industrial gas turbines for power generation are advanced engines that require constant and detailed monitorization using internal and external sensors. These sensors generate a large flow of data in the form of multivariate time series that are amenable to analysis using pattern recognition methods with the objective of improving and optimizing turbine operation. One aspect this may take is visual analytics, where dimensionality reduction methods can be used to intuitively visualize the multivariate time series. This brief paper provides a proof of concept and some case scenarios of a visual turbine-monitorization tool, based on the UMAP method, combined with clustering using HDBSCAN and unsupervised Agnostic Feature Selection. It can be considered as a first step towards a data-centered approach to gas turbine management within the industry 4.0 framework, based on the mining of turbine sensor data.

Machine learning-based systems are now part of a wide array of real-world applications seamlessly embedded in the social realm. In the wake of this realisation, strict legal regulations for these systems are currently being developed, addressing some of the risks they may pose. This is the coming of age of the interpretability and explainability problems in machine learning-based data analysis, which can no longer be seen just as an academic research problem. In this tutorial, associated to ESANN 2021 special session on “Interpretable Models in Machine Learning and Explainable Artificial Intelligence”, we discuss explainable and interpretable machine learning as post-hoc and ante-hoc strategies to address these problems and highlight several aspects related to them, including their assessment. The contributions accepted for the session are then presented in this context

Background In-vivo single voxel proton magnetic resonance spectroscopy (SV 1H-MRS), coupled with supervised pattern recognition (PR) methods, has been widely used in clinical studies of discrimination of brain tumour types and follow-up of patients bearing abnormal brain masses. SV 1H-MRS provides useful biochemical information about the metabolic state of tumours and can be performed at short (< 45 ms) or long (> 45 ms) echo time (TE), each with particular advantages. Short-TE spectra are more adequate for detecting lipids, while the long-TE provides a much flatter signal baseline in between peaks but also negative signals for metabolites such as lactate. Both, lipids and lactate, are respectively indicative of specific metabolic processes taking place. Ideally, the information provided by both TE should be of use for clinical purposes. In this study, we characterise the performance of a range of Non-negative Matrix Factorisation (NMF) methods in two respects: first, to derive sources correlated with the mean spectra of known tissue types (tumours and normal tissue); second, taking the best performing NMF method for source separation, we compare its accuracy for class assignment when using the mixing matrix directly as a basis for classification, as against using the method for dimensionality reduction (DR). For this, we used SV 1H-MRS data with positive and negative peaks, from a widely tested SV 1H-MRS human brain tumour database. Results The results reported in this paper reveal the advantage of using a recently described variant of NMF, namely Convex-NMF, as an unsupervised method of source extraction from SV1H-MRS. Most of the sources extracted in our experiments closely correspond to the mean spectra of some of the analysed tumour types. This similarity allows accurate diagnostic predictions to be made both in fully unsupervised mode and using Convex-NMF as a DR step previous to standard supervised classification. The obtained results are comparable to, or more accurate than those obtained with supervised techniques. Conclusions The unsupervised properties of Convex-NMF place this approach one step ahead of classical label-requiring supervised methods for the discrimination of brain tumour types, as it accounts for their increasingly recognised molecular subtype heterogeneity. The application of Convex-NMF in computer assisted decision support systems is expected to facilitate further improvements in the uptake of MRS-derived information by clinicians.

Glioblastoma is the most frequent aggressive primary brain tumor amongst human adults. Its standard treatment involves chemotherapy, for which the drug temozolomide is a common choice. These are heterogeneous and variable tumors which might benefit from personalized, data-based therapy strategies, and for which there is room for improvement in therapy response follow-up, investigated with preclinical models. This study addresses a preclinical question that involves distinguishing between treated and control (untreated) mice bearing glioblastoma, using machine learning techniques, from magnetic resonance-based data in two modalities: MRI and MRSI. It aims to go beyond the comparison of methods for such discrimination to provide an analytical pipeline that could be used in subsequent human studies. This analytical pipeline is meant to be a usable and interpretable tool for the radiology expert in the hope that such interpretation helps revealing new insights about the problem itself. For that, we propose coupling source extraction-based and radiomics-based data transformations with feature selection. Special attention is paid to the generation of radiologist-friendly visual nosological representations of the analyzed tumors. L.M.N., A.V., C.A., A.P.C., M.J.L. and A.S. received EC funding from the ATTRACT project, under Grant Agreement 777222. A.V. and E.R. acknowledge funding from Spanish MINECO TIN2016-79576-R research project. A.S and M.J.L-C acknowledge funding from Spanish MINECO RTI2018-098682-B-I00.

Background: The 2019 Science for Dialysis Meeting at Bellvitge University Hospital was devoted to the challenges and opportunities posed by the use of data science to facilitate precision and personalized medicine in nephrology, and to describe new approaches and technologies. The meeting included separate sections for issues in data collection and data analysis. As part of data collection, we presented the institutional ARGOS e-health project, which provides a common model for the standardization of clinical practice. We also pay specific attention to the way in which randomized controlled trials offer data that may be critical to decision-making in the real world. The opportunities of open source software (OSS) for data science in clinical practice were also discussed. Summary: Precision medicine aims to provide the right treatment for the right patients at the right time and is deeply connected to data science. Dialysis patients are highly dependent on technology to live, and their treatment generates a huge volume of data that has to be analysed. Data science has emerged as a tool to provide an integrated approach to data collection, storage, cleaning, processing, analysis, and interpretation from potentially large volumes of information. This is meant to be a perspective article about data science based on the experience of the experts invited to the Science for Dialysis Meeting and provides an up-to-date perspective of the potential of data science in kidney disease and dialysis. Key messages: Healthcare is quickly becoming data-dependent, and data science is a discipline that holds the promise of contributing to the development of personalized medicine, although nephrology still lags behind in this process. The key idea is to ensure that data will guide medical decisions based on individual patient characteristics rather than on averages over a whole population usually based on randomized controlled trials that excluded kidney disease patients. Furthermore, there is increasing interest in obtaining data about the effectiveness of available treatments in current patient care based on pragmatic clinical trials. The use of data science in this context is becoming increasingly feasible in part thanks to the swift developments in OSS.

Machine learning (ML) methods have shown great potential for the analysis of data involved in medical decisions. However, for these methods to be incorpored in the medical pipeline, they must be made interpretable not only to the data analyst, but also to the medical expert. In this work, we have applied a combination of feature transformation, selection and classification using ML and statistical methods to differentiate between control (untreated) and Temozolomide (TMZ)-treated tumour tissue from a glioblastoma (brain tumour) murine model. As input, we have used T2 weighted magnetic resonance images (MRI) and spectroscopic imaging (MRSI). Radiomics features have been extracted from the MRI dataset, while convex Non-negative Matrix Factorization (Convex-NMF) was used to extract sources from the MRSI dataset. Exhaustive feature selection has revealed parsimonious feature subsets that facilitate the expert interpretation of results while retaining a high discriminatory ability. L.M.N., A.V., C.A. and A.P.C. received EC funding from the ATTRACT project, under Grant Agreement 777222. A.V. and E.R. acknowledge funding from Spanish MINECO TIN2016- 79576-R research project.

Background: Like other scientific fields, such as cosmology, high-energy physics, or even the life sciences, medicine and healthcare face the challenge of an extremely quick transformation into data-driven sciences. This challenge entails the daunting task of extracting usable knowledge from these data using algorithmic methods. In the medical context this may for instance realized through the design of medical decision support systems for diagnosis, prognosis and patient management. The intensive care unit (ICU), and by extension the whole area of critical care, is becoming one of the most data-driven clinical environments. Results: The increasing availability of complex and heterogeneous data at the point of patient attention in critical care environments makes the development of fresh approaches to data analysis almost compulsory. Computational Intelligence (CI) and Machine Learning (ML) methods can provide such approaches and have already shown their usefulness in addressing problems in this context. The current study has a dual goal: it is first a review of the state-of-the-art on the use and application of such methods in the field of critical care. Such review is presented from the viewpoint of the different subfields of critical care, but also from the viewpoint of the different available ML and CI techniques. The second goal is presenting a collection of results that illustrate the breath of possibilities opened by ML and CI methods using a single problem, the investigation of septic shock at the ICU. Conclusion: We have presented a structured state-of-the-art that illustrates the broad-ranging ways in which ML and CI methods can make a difference in problems affecting the manifold areas of critical care. The potential of ML and CI has been illustrated in detail through an example concerning the sepsis pathology. The new definitions of sepsis and the relevance of using the systemic inflammatory response syndrome (SIRS) in its diagnosis have been considered. Conditional independence models have been used to address this problem, showing that SIRS depends on both organ dysfunction measured through the Sequential Organ Failure (SOFA) score and the ICU outcome, thus concluding that SIRS should still be considered in the study of the pathophysiology of Sepsis. Current assessment of the risk of dead at the ICU lacks specificity. ML and CI techniques are shown to improve the assessment using both indicators already in place and other clinical variables that are routinely measured. Kernel methods in particular are shown to provide the best performance balance while being amenable to representation through graphical models, which increases their interpretability and, with it, their likelihood to be accepted in medical practice.

Despite the success of automated pattern recognition methods in problems of human brain tumor diagnostic classification, limited attention has been paid to the issue of automated data quality assessment in the field of MRS for neuro-oncology. Beyond some early attempts to address this issue, the current standard in practice is MRS quality control through human (expert-based) assessment. One aspect of automatic quality control is the problem of detecting artefacts in MRS data. Artefacts, whose variety has already been reviewed in some detail and some of which may even escape human quality control, have a negative influence in pattern recognition methods attempting to assist tumor characterization. The automatic detection of MRS artefacts should be beneficial for radiology as it guarantees more reliable tumor characterizations, as well as the development of more robust pattern recognition-based tumor classifiers and more trustable MRS data processing and analysis pipelines. Feature extraction methods have previously been used to help distinguishing between good and bad quality spectra to apply subsequent supervised pattern recognition techniques. In this study, we apply feature extraction differently and use a variant of a method for blind source separation, namely Convex Non-Negative Matrix Factorization, to unveil MRS signal sources in a completely unsupervised way. We hypothesize that, while most sources will correspond to the different tumor patterns, some of them will reflect signal artefacts. The experimental work reported in this paper, analyzing a combined short and long echo time 1H-MRS database of more than 2000 spectra acquired at 1.5T and corresponding to different tumor types and other anomalous masses, provides a first proof of concept that points to the possible validity of this approach.

The use of machine learning (ML) approaches to target clinical problems is called to revolutionize clinical decision-making. The success of these tools is subjected to the understanding of the intrinsic processes being used during the classical pathway by which clinicians make decisions. In a parallelism with this pathway, ML can have an impact at four levels: for data acquisition, predominantly by extracting standardized, high-quality information with the smallest possible learning curve; for feature extraction, by discharging healthcare practitioners from performing tedious measurements on raw data; for interpretation, by digesting complex, heterogeneous data in order to augment the understanding of the patient status; and for decision support, by leveraging the previous step to predict clinical outcomes, response to treatment or to recommend a specific intervention. This paper discusses the state-of-the-art, as well as the current clinical status and challenges associated with each of these tasks, together with the challenges related to the learning process, the auditability/traceability, the system infrastructure and the integration within clinical processes.

Background: Modern clinical environments are laden with technology devices continuously gathering physiological data from patients. This is especially true in critical care environments, where life-saving decisions may have to be made on the basis of signals from monitoring devices. Hemodynamic monitoring is essential in dialysis, surgery, and in critically ill patients. For the most severe patients, blood pressure is normally assessed through a catheter, which is an invasive procedure that may result in adverse effects. Blood pressure can also be monitored noninvasively through different methods and these data can be used for the continuous assessment of pressure using machine learning methods. Previous studies have found pulse transit time to be related to blood pressure. In this short paper, we propose to study the feasibility of implementing a data-driven model based on restricted Boltzmann machine artificial neural networks, delivering a first proof of concept for the validity and viability of a method for blood pressure prediction based on these models. Summary and Key Messages: For the most severe patients (e.g., dialysis, surgery, and the critically ill), blood pressure is normally assessed through invasive catheters. Alternatively, noninvasive methods have also been developed for its monitorization. Data obtained from noninvasive measurements can be used for the continuous assessment of pressure using machine learning methods. In this study, a restricted Boltzmann machine artificial neural network is used to present a first proof of concept for the validity and viability of a method for blood pressure prediction.

In a short period of time, many areas of science have made a sharp transition towards data-dependent methods. In some cases, this process has been enabled by simultaneous advances in data acquisition and the development of networked system technologies. This new situation is particularly clear in the life sciences, where data overabundance has sparked a flurry of new methodologies for data management and analysis. This can be seen as a perfect scenario for the use of machine learning and computational intelligence techniques to address problems in which more traditional data analysis approaches might struggle. But, this scenario also poses some serious challenges. One of them is model interpretability and explainability, especially for complex nonlinear models. In some areas such as medicine and health care, not addressing such challenge might seriously limit the chances of adoption, in real practice, of computer-based systems that rely on machine learning and computational intelligence methods for data analysis. In this paper, we reflect on recent investigations about the interpretability and explainability of machine learning methods and discuss their impact on medicine and health care. We pay specific attention to one of the ways in which interpretability and explainability in this context can be addressed, which is through data and model visualization. We argue that, beyond improving model interpretability as a goal in itself, we need to integrate the medical experts in the design of data analysis interpretation strategies. Otherwise, machine learning is unlikely to become a part of routine clinical and health care practice.

Background: Medicine is becoming an increasingly data-centred discipline and, beyond classical statistical approaches, artificial intelligence (AI) and, in particular, machine learning (ML) are attracting much interest for the analysis of medical data. It has been argued that AI is experiencing a fast process of commodification. This characterization correctly reflects the current process of industrialization of AI and its reach into society. Therefore, societal issues related to the use of AI and ML should not be ignored any longer and certainly not in the medical domain. These societal issues may take many forms, but they all entail the design of models from a human-centred perspective, incorporating human-relevant requirements and constraints. In this brief paper, we discuss a number of specific issues affecting the use of AI and ML in medicine, such as fairness, privacy and anonymity, explainability and interpretability, but also some broader societal issues, such as ethics and legislation. We reckon that all of these are relevant aspects to consider in order to achieve the objective of fostering acceptance of AI- and ML-based technologies, as well as to comply with an evolving legislation concerning the impact of digital technologies on ethically and privacy sensitive matters. Our specific goal here is to reflect on how all these topics affect medical applications of AI and ML. This paper includes some of the contents of the “2nd Meeting of Science and Dialysis: Artificial Intelligence,” organized in the Bellvitge University Hospital, Barcelona, Spain. Summary and Key Messages: AI and ML are attracting much interest from the medical community as key approaches to knowledge extraction from data. These approaches are increasingly colonizing ambits of social impact, such as medicine and healthcare. Issues of social relevance with an impact on medicine and healthcare include (although they are not limited to) fairness, explainability, privacy, ethics and legislation.

G-protein-coupled receptors (GPCRs) are a large and diverse super-family of eukaryotic cell membrane proteins that play an important physiological role as transmitters of extracellular signal. In this paper, we investigate Class C, a member of this super-family that has attracted much attention in pharmacology. The limited knowledge about the complete 3D crystal structure of Class C receptors makes necessary the use of their primary amino acid sequences for analytical purposes. Here, we provide a systematic analysis of distinct receptor sequence segments with regard to their ability to differentiate between seven class C GPCR subtypes according to their topological location in the extracellular, transmembrane, or intracellular domains. We build on the results from the previous research that provided preliminary evidence of the potential use of separated domains of complete class C GPCR sequences as the basis for subtype classification. The use of the extracellular N-terminus domain alone was shown to result in a minor decrease in subtype discrimination in comparison with the complete sequence, despite discarding much of the sequence information. In this paper, we describe the use of Support Vector Machine-based classification models to evaluate the subtype-discriminating capacity of the specific topological sequence segments.

Biocuration in the omics sciences has become paramount, as research in these fields rapidly evolves towards increasingly data-dependent models. As a result, the management of web-accessible publicly-available databases becomes a central task in biological knowledge dissemination. One relevant challenge for biocurators is the unambiguous identification of biological entities. In this study, we illustrate the adequacy of machine learning methods as biocuration assistance tools using a publicly available protein database as an example. This database contains information on G Protein-Coupled Receptors (GPCRs), which are part of eukaryotic cell membranes and relevant in cell communication as well as major drug targets in pharmacology. These receptors are characterized according to subtype labels. Previous analysis of this database provided evidence that some of the receptor sequences could be affected by a case of label noise, as they appeared to be too consistently misclassified by machine learning methods. Here, we extend our analysis to recent and quite substantially modified new versions of the database and reveal their now extremely accurate labeling using several machine learning models and different transformations of the unaligned sequences. These findings support the adequacy of our proposed method to identify problematic labeling cases as a tool for database biocuration.

Background: Current dialysis devices are not able to react when unexpected changes occur during dialysis treatment or to learn about experience for therapy personalization. Furthermore, great efforts are dedicated to develop miniaturized artificial kidneys to achieve a continuous and personalized dialysis therapy, in order to improve the patient’s quality of life. These innovative dialysis devices will require a real-time monitoring of equipment alarms, dialysis parameters, and patient-related data to ensure patient safety and to allow instantaneous changes of the dialysis prescription for the assessment of their adequacy. The analysis and evaluation of the resulting large-scale data sets enters the realm of “big data” and will require real-time predictive models. These may come from the fields of machine learning and computational intelligence, both included in artificial intelligence, a branch of engineering involved with the creation of devices that simulate intelligent behavior. The incorporation of artificial intelligence should provide a fully new approach to data analysis, enabling future advances in personalized dialysis therapies. With the purpose to learn about the present and potential future impact on medicine from experts in artificial intelligence and machine learning, a scientific meeting was organized in the Hospital Universitari Bellvitge (L’Hospitalet, Barcelona). As an outcome of that meeting, the aim of this review is to investigate artificial intel ligence experiences on dialysis, with a focus on potential barriers, challenges, and prospects for future applications of these technologies. Summary and Key Messages: Artificial intelligence research on dialysis is still in an early stage, and the main challenge relies on interpretability and/or comprehensibility of data models when applied to decision making. Artificial neural networks and medical decision support systems have been used to make predictions about anemia, total body water, or intradialysis hypotension and are promising approaches for the prescription and monitoring of hemodialysis therapy. Current dialysis machines are continuously improving due to innovative technological developments, but patient safety is still a key challenge. Real-time monitoring systems, coupled with automatic instantaneous biofeedback, will allow changing dialysis prescriptions continuously. The integration of vital sign monitoring with dialysis parameters will produce large data sets that will require the use of data analysis techniques, possibly from the area of machine learning, in order to make better decisions and increase the safety of patients.

G protein-coupled receptors are eukaryotic cell membrane proteins with a key role as extracellular signal transmitters. While GPCRs embrace a wide and heterogeneous super-family of proteins, our interest in this study is in its Class C, of great relevance to pharmacology. The scarcity of knowledge about their full 3-D crystal structure makes the use of their primary amino acid sequences important for analysis. In this paper, we systematically analyze whether segments of the receptor sequences are able to discriminate between the different class C GPCR subtypes according to their topological location on the extracellular, transmembrane or intracellular domain. For this, we build on previous research that showed that the use of the extracellular N-terminus domain on its own for this classification task did only entail a minor decrease in subtype discrimination when compared to the complete sequence. We use Support Vector Machine-based classification models to assess the subtype discriminating power of the topological segments.

The current European legislation requires households the installation of smart metering systems. These will eventually allow electric utilities to gather richly detailed data of consumption. In this scenario, the implementation of data mining procedures for actionable knowledge extraction could be the key to competitive advantage. These may take the form of market segmentation using clustering techniques for the identification of customer behaviour patterns of electricity consumption that could justify the definition of tailored tariffs. In this brief paper, we show that the combination of a standard clustering algorithm with a similarity measure specifically defined for non-i.i.d. data, namely Dynamic Time Warping, can reveal an actionable segmentation of a real consumer market, combining business criteria and quantitative evaluation.

G protein-coupled receptors (GPCRs) are integral cell membrane proteins of relevance for pharmacology. The tertiary structure of the transmembrane domain, a gate to the study of protein functionality, is unknown for almost all members of class C GPCRs, which are the target of the current study. As a result, their investigation must often rely on alignments of their amino acid sequences. Sequence alignment entails the risk of missing relevant information. Various approaches have attempted to circumvent this risk through alignment-free transformations of the sequences on the basis of different amino acid physicochemical properties. In this paper, we use several of these alignment-free methods, as well as a basic amino acid composition representation, to transform the available sequences. Novel semi-supervised statistical machine learning methods are then used to discriminate the different class C GPCRs types from the transformed data. This approach is relevant due to the existence of orphan proteins to which type labels should be assigned in a process of deorphanization or reverse pharmacology. The reported experiments show that the proposed techniques provide accurate classification even in settings of extreme class-label scarcity and that fair accuracy can be achieved even with very simple transformation strategies that ignore the sequence ordering.

Contemporary medical science heavily relies on the use of technology. Part of this technology strives to im- prove examination and measurement of the human body, with some of the most impressive technical breakthroughs to be found in the development of non-invasive proce- dures. Another part focuses on the development of de- vices that support therapies for specific pathologies. An example of this is the artificial kidney, which has become the target of intensive research from many directions and generates great expectations for dialysis patients. Re- search on the artificial kidney is still incipient though, and there are many challenges that must be overcome before it will become a reality and part of clinical practice in ne- phrology. One of these non-trivial challenges concerns the safety of users of these new dialysis devices. Safety risks make effective monitoring systems mandatory.

Real-world applications of multivariate data analysis often stumble upon the barrier of interpretability. Simple data analysis methods are usually easy to interpret, but they risk providing poor data models. More involved methods may instead yield faithful data models, but limited interpretability. This is the case of linear and nonlinear methods for multivariate data visualization through dimensionality reduction. Even though the latter have provided some of the most exciting visualization developments, their practicality is hindered by the difficulty of explaining them in an intuitive manner. The interpretability, and therefore the practical applicability, of data visualization through nonlinear dimensionality reduction (NLDR) methods would improve if, first, we could accurately calculate the distortion introduced by these methods in the visual representation and, second, if we could faithfully reintroduce this distortion into such representation. In this paper, we describe a technique for the reintroduction of the distortion into the visualization space of NLDR models. It is based on the concept of density-equalizing maps, or cartograms, recently developed for the representation of geographic information. We illustrate it using Generative Topographic Mapping (GTM), a nonlinear manifold learning method that can provide both multivariate data visualization and a measure of the local distortion that the model generates. Although illustrated here with GTM, it could easily be extended to other NLDR visualization methods, provided a local distortion measure could be calculated. It could also serve as a guiding tool for interactive data visualization.

G protein-coupled receptors are a large and heterogeneous super-family of cell membrane proteins of interest to biology in general. One of its families, class C, is of particular interest to pharmacology and drug design. This family is quite heterogeneous on its own, and the discrimination of its several sub-families is a challenging problem. In the absence of known crystal structure, such discrimination must rely on their primary amino acid sequences. In this study, we are interested not as much in achieving maximum sub-family discrimination accuracy, but in exploring sequence misclassification behaviour. Specifically, we are interested in isolating those sequences showing consistent misclassification, that is, sequences that are very often misclassified and almost always to the same wrong sub-family. This analysis should assist database curators in receptor quality control tasks. Random Forests are used for this analysis due their ensemble nature, which makes them naturally suited to gauge the consistency of misclassification.

The diagnosis of brain tumours is an extremely sensitive and complex clinical task that must rely upon information gathered through non-invasive techniques. One such technique is Magnetic Resonance Spectroscopy. In this task, radiology experts are likely to benefit from the support of computer-based systems built around robust classification processes. In this paper, a Discrete Wavelet Transform procedure was applied to the pre-processing of spectra corresponding to several brain tumour pathologies. This procedure does not alleviate the high dimensionality of the data by itself. For this reason, dimensionality reduction was subsequently implemented using Moving Window with Variance Analysis for feature selection or Principal Component Analysis for feature extraction. The combined method yielded very encouraging results in terms of diagnostic discriminatory binary classification using Bayesian Neural Networks. In most cases, the classification accuracy improved on previously reported results.

This article investigates methods for the accurate and robust differentiation of metastases from glioblastomas on the basis of single-voxel 1H MRS information. Single-voxel 1H MR spectra from a total of 109 patients (78 glioblastomas and 31 metastases) from the multicenter, international INTERPRET database, plus a test set of 40 patients (30 glioblastomas and 10 metastases) from three different centers in the Barcelona (Spain) metropolitan area, were analyzed using a robust method for feature (spectral frequency) selection coupled with a linear-in-the-parameters single-layer perceptron classifier. For the test set, a parsimonious selection of five frequencies yielded an area under the receiver operating characteristic curve of 0.86, and an area under the convex hull of the receiver operating characteristic curve of 0.91. Moreover, these accurate results for the discrimination between glioblastomas and metastases were obtained using a small number of frequencies that are amenable to metabolic interpretation, which should ease their use as diagnostic markers. Importantly, the prediction can be expressed as a simple formula based on a linear combination of these frequencies. As a result, new cases could be straightforwardly predicted by integrating this formula into a computer-based medical decision support system. This work also shows that the combination of spectra acquired at different TEs (short TE, 20–32 ms; long TE, 135–144 ms) is key to the successful discrimination between glioblastomas and metastases from single-voxel 1H MRS. Copyright © 2011 John Wiley & Sons, Ltd.

Transcranial magnetic stimulation (TMS) is a powerful tool for the calculation of parameters related to the intracortical excitability and inhibition of the motor cortex. The cortical silent period (CSP) is one such parameter that corresponds to the suppression of muscle activity for a short period after a muscle response to TMS. The duration of the CSP is known to be correlated with the prognosis of brain stroke patients' motor ability. Current methods for the estimation of the CSP duration are very sensitive to the presence of noise. A variational Bayesian formulation of a manifold-constrained hidden Markov model is applied in this paper to the segmentation of a set of multivariate time series (MTS) of electromyographic recordings corresponding to stroke patients and control subjects. A novel index of variability associated to this model is defined and applied to the detection of the silent period interval of the signal and to the estimation of its duration. This model and its associated index are shown to behave robustly in the presence of noise and provide more reliable estimations than the current standard in clinical practice.

We present a novel semi-supervised model, SS-Geo-GTM, which stems from a geodesic distance-based extension of Generative Topographic Mapping that prioritizes neighbourhood relationships along a generated manifold embedded in the observed data space. With this, it improves the trustworthiness and the continuity of the low-dimensional representations it provides, while behaving robustly in the presence of noise. In SS-Geo-GTM, the model prototypes are linked by the nearest neighbour to the data manifold constructed by Geo-GTM. The resulting proximity graph is used as the basis for a class label propagation algorithm. The performance of SS-Geo-GTM is experimentally assessed, comparing positively with that of an Euclidean distance-based counterpart and with those of alternative manifold learning methods.

Machine Learning (ML) and related methods have of late made significant contributions to solving multidisciplinary problems in the field of oncology diagnosis. Human brain tumor diagnosis, in particular, often relies on the use of non-invasive techniques such as Magnetic Resonance Imaging (MRI) and Spectroscopy (MRS). In this paper, MRS data of human brain tumors are analyzed in detail. The high dimensionality of the MR spectra makes difficult both their classification and the interpretation of the obtained results, thus limiting their usability in practical medical settings. The use of dimensionality reduction techniques is therefore advisable. In this work, we apply feature selection methods and several off-the-shelf classifiers on various ^1H-MRS modalities: long and short echo times and an ad hoc combination of both. The introduction of bootstrap resampling techniques permits the obtention of mean performance estimates and their variability. Our experimental findings indicate that the feature selection process enhances the classification performance compared to using the full set of features. We also show that the use of combined information from the different echo times is a better strategy for small numbers of spectral frequencies; however, the use of ever greater numbers of short echo time frequencies permits the obtention of many models with similar performance. The final induced models offer very attractive solutions both in terms of prediction accuracy and number of involved spectral frequencies, which are also amenable to metabolic interpretation. A linear dimensionality-reduction technique that preserves class discrimination capabilities is used for visualizing the data corresponding to the selected frequencies.

General finite mixture models are powerful tools for the density-based grouping of multivariate i.i.d. data, but they lack data visualization capabilities, which reduces their practical applicability to real-world problems. Generative topographic mapping (GTM) was originally formulated as a constrained mixture of distributions in order to provide simultaneous visualization and clustering of multivariate data. In its inception, the adaptive parameters were determined by maximum likelihood (ML), using the expectation-maximization (EM) algorithm. The original GTM is, therefore, prone to data overfitting unless a regularization mechanism is included. In this paper, we define an alternative variational formulation of GTM that provides a full Bayesian treatment to a Gaussian process (GP)-based variation of the model. The generalization capabilities of the proposed Variational Bayesian GTM are assessed in some detail and compared with those of alternative GTM regularization approaches in terms of test log-likelihood, using several artificial and real datasets.

The investigation of protein functionality often relies on the knowledge of crystal 3-D structure. This structure is not always known or easily unravelled, which is the case of eukaryotic cell membrane proteins such as G Protein-Coupled Receptors (GPCRs) and specially of those of class C, which are the target of the current study. In the absence of information about tertiary or quaternary structures, functionality can be investigated from the primary structure, that is, from the amino acid sequence. In previous research, we found that the different subtypes of class C GPCRs could be discriminated with a high level of accuracy from the n-gram transformation of their complete primary sequences, using a method that combined two-stage feature selection with kernel classifiers. This study aims at discovering whether subunits of the complete sequence retain such discrimination capabilities. We report experiments that show that the extracellular N-terminal domain of the receptor suffices to retain the classification accuracy of the complete sequence and that it does so using a reduced selection of n-grams whose length of up to five amino acids opens up an avenue for class C GPCR signature motif discovery.

Uncertainty is inherent in medical decision making and poses a challenge for intelligent technologies. This paper focuses on magnetic resonance spectra (MRS) for discrimination of brain tumour types and grades. Modelling of this type of high-dimensional data is commonly affected by uncertainty caused by the presence of outliers. Multivariate data clustering and visualization of MRS data is proposed using the GTM framework with basis functions comprising Student t-distributions in order to minimize the negative impact on the model from outliers. The effectiveness of this model on the MRS data is demonstrated empirically.

This paper proposes a principled, self-organized, framework to manage two sources of uncertainty that are inherent in intelligent systems for medical decision support, namely outliers and missing data. The framework is applied to magnetic resonance spectra (MRS), which are indicators of the grade of malignancy in brain tumours. A model for multivariate data clustering and visualization, the generative topographic mapping (GTM), is re-formulated as a mixture of Student's t-distributions making it more robust to outliers while supporting the imputation of missing values. An important new development is the extension of the model to provide automatic feature relevance determination. Its effectiveness on the MRS data is demonstrated empirically.

Neuro-oncologists must ultimately rely on their acquired knowledge and accumulated experience to undertake the sensitive task of brain tumour diagnosis. This task strongly depends on indirect, non-invasive measurements, which are the source of valuable data in the form of signals and images. Expert radiologists should benefit from their use as part of an at least partially automated computer-based medical decision support system. This paper focuses on Magnetic Resonance Spectroscopy signal analysis and illustrates a method that combines Gaussian Decomposition, dimensionality reduction by Moving Window with Variance Analysis and classification using adaptively regularized Artificial Neural Networks. The method yields encouraging results in the task of binary classification of human brain tumours, even for tumour types that have seldom been analyzed from this viewpoint.

Objective: This paper presents an algorithm to assess the risk of death in patients with sepsis. Sepsis is a common clinical syndrome in the intensive care unit (ICU) that can lead to severe sepsis, a severe state of septic shock or multi-organ failure. The proposed algorithm may be implemented as part of a clinical decision support system that can be used in combination with the scores deployed in the ICU to improve the accuracy, sensitivity and specificity of mortality prediction for patients with sepsis. Methodology: In this paper, we used the Simplified Acute Physiology Score (SAPS) for ICU patients and the Sequential Organ Failure Assessment (SOFA) to build our kernels and algorithms. In the proposed method, we embed the available data in a suitable feature space and use algorithms based on linear algebra, geometry and statistics for inference. We present a simplified version of the Fisher kernel (practical Fisher kernel for multinomial distributions), as well as a novel kernel that we named the Quotient Basis Kernel (QBK). These kernels are used as the basis for mortality prediction using soft-margin support vector machines. The two new kernels presented are compared against other generative kernels based on the Jensen–Shannon metric (centred, exponential and inverse) and other widely used kernels (linear, polynomial and Gaussian). Clinical relevance is also evaluated by comparing these results with logistic regression and the standard clinical prediction method based on the initial SAPS score. Results: As described in this paper, we tested the new methods via cross-validation with a cohort of 400 test patients. The results obtained using our methods compare favourably with those obtained using alternative kernels (80.18% accuracy for the QBK) and the standard clinical prediction method, which are based on the basal SAPS score or logistic regression (71.32% and 71.55%, respectively). The QBK presented a sensitivity and specificity of 79.34% and 83.24%, which outperformed the other kernels analysed, logistic regression and the standard clinical prediction method based on the basal SAPS score. Conclusion: Several scoring systems for patients with sepsis have been introduced and developed over the last 30 years. They allow for the assessment of the severity of disease and provide an estimate of in-hospital mortality. Physiology-based scoring systems are applied to critically ill patients and have a number of advantages over diagnosis-based systems. Severity score systems are often used to stratify critically ill patients for possible inclusion in clinical trials. In this paper, we present an effective algorithm that combines both scoring methodologies for the assessment of death in patients with sepsis that can be used to improve the sensitivity and specificity of the currently available methods.

Convex non-negative matrix factorization is a blind signal separation technique that has previously demonstrated to be well-suited for the task of human brain tumor diagnosis from magnetic resonance spectroscopy data. This is due to its ability to retrieve interpretable sources of mixed sign that highly correlate with tissue type prototypes. The current study provides a Bayesian formulation for such problem and derives a maximum a posteriori estimate based on a gradient descent algorithm specifically designed to deal with matrices with different sign restrictions. Its applicability to neuro-oncology diagnosis was experimentally assessed and the results were found to be comparable to those achieved by state of the art methods in tumor type discrimination and consistently better in source extraction.