RESUMEN

The process of developing new drugs is widely acknowledged as being time-intensive and requiring substantial financial investment. Despite ongoing efforts to reduce time and expenses in drug development, ensuring medication safety remains an urgent problem. One of the major problems involved in drug development is hepatotoxicity, specifically known as drug-induced liver injury (DILI). The popularity of new drugs often poses a significant barrier during development and frequently leads to their recall after launch. In silico methods have many advantages compared with traditional in vivo and in vitro assays. To establish a more precise and reliable prediction model, it is necessary to utilize an extensive and high-quality database consisting of information on drug molecule properties and structural patterns. In addition, we should also carefully select appropriate molecular descriptors that can be used to accurately depict compound characteristics. The aim of this study was to conduct a comprehensive investigation into the prediction of DILI. First, we conducted a comparative analysis of the physicochemical properties of extensively well-prepared DILI-positive and DILI-negative compounds. Then, we used classic substructure dissection methods to identify structural pattern differences between these two different types of chemical molecules. These findings indicate that it is not feasible to establish property or substructure-based rules for distinguishing between DILI-positive and DILI-negative compounds. Finally, we developed quantitative classification models for predicting DILI using the naïve Bayes classifier (NBC) and recursive partitioning (RP) machine learning techniques. The optimal DILI prediction model was obtained using NBC, which combines 21 physicochemical properties, the VolSurf descriptors and the LCFP_10 fingerprint set. This model achieved a global accuracy (GA) of 0.855 and an area under the curve (AUC) of 0.704 for the training set, while the corresponding values were 0.619 and 0.674 for the test set, respectively. Moreover, indicative substructural fragments favorable or unfavorable for DILI were identified from the best naïve Bayesian classification model. These findings may help prioritize lead compounds in the early stage of drug development pipelines.

Asunto(s)

RESUMEN

PURPOSE: We built Bayesian Network (BN) models to explain roles of different patient-specific factors affecting racial differences in breast cancer stage at diagnosis, and to identify healthcare related factors that can be intervened to reduce racial health disparities. METHODS: We studied women age 67-74 with initial diagnosis of breast cancer during 2006-2014 in the National Cancer Institute's SEER-Medicare dataset. Our models included four measured variables (tumor grade, hormone receptor status, screening utilization and biopsy delay) expressed through two latent pathways-a tumor biology path, and health-care access/utilization path. We used various Bayesian model assessment tools to evaluate these two latent pathways as well as each of the four measured variables in explaining racial disparities in stage-at-diagnosis. RESULTS: Among 3,010 Black non-Hispanic (NH) and 30,310 White NH breast cancer patients, respectively 70.2% vs 76.9% were initially diagnosed at local stage, 25.3% vs 20.3% with regional stage, and 4.56% vs 2.80% with distant stage-at-diagnosis. Overall, BN performed approximately 4.7 times better than Classification And Regression Tree (CART) (Breiman L, Friedman JH, Stone CJ, Olshen RA. Classification and regression trees. CRC press; 1984) in predicting stage-at-diagnosis. The utilization of screening mammography is the most prominent contributor to the accuracy of the BN model. Hormone receptor (HR) status and tumor grade are useful for explaining racial disparity in stage-at diagnosis, while log-delay in biopsy impeded good prediction. CONCLUSIONS: Mammography utilization had a significant effect on racial differences in breast cancer stage-at-diagnosis, while tumor biology factors had less impact. Biopsy delay also aided in predicting local and regional stages-at-diagnosis for Black NH women but not for white NH women.

Asunto(s)

RESUMEN

OBJECTIVE: To predict the influencing factors of neonatal pneumonia in pregnant women with diabetes mellitus using a Bayesian network model. By examining the intricate network connections between the numerous variables given by Bayesian networks (BN), this study aims to compare the prediction effect of the Bayesian network model and to analyze the influencing factors directly associated to neonatal pneumonia. METHOD: Through the structure learning algorithms of BN, Naive Bayesian (NB), Tree Augmented Naive Bayes (TAN), and k-Dependence Bayesian Classifier (KDB), complex networks connecting variables were presented and their predictive abilities were tested. The BN model and three machine learning models computed using the R bnlean package were also compared in the data set. RESULTS: In constraint-based algorithms, three algorithms had different presentation DAGs. KDB had a better prediction effect than NB and TAN, and it achieved higher AUC compared with TAN. Among three machine learning modes, Support Vector Machine showed a accuracy rate of 91.04% and 67.88% of precision, which was lower than TAN (92.70%; 72.10%). CONCLUSION: KDB was applicable, and it can detect the dependencies between variables, identify more potential associations and track changes between variables and outcome.

Asunto(s)

RESUMEN

BACKGROUND: The Naive Bayes (NB) classifier is a powerful supervised algorithm widely used in Machine Learning (ML). However, its effectiveness relies on a strict assumption of conditional independence, which is often violated in real-world scenarios. To address this limitation, various studies have explored extensions of NB that tackle the issue of non-conditional independence in the data. These approaches can be broadly categorized into two main categories: feature selection and structure expansion. In this particular study, we propose a novel approach to enhancing NB by introducing a latent variable as the parent of the attributes. We define this latent variable using a flexible technique called Bayesian Latent Class Analysis (BLCA). As a result, our final model combines the strengths of NB and BLCA, giving rise to what we refer to as NB-BLCA. By incorporating the latent variable, we aim to capture complex dependencies among the attributes and improve the overall performance of the classifier. METHODS: Both Expectation-Maximization (EM) algorithm and the Gibbs sampling approach were offered for parameter learning. A simulation study was conducted to evaluate the classification of the model in comparison with the ordinary NB model. In addition, real-world data related to 976 Gastric Cancer (GC) and 1189 Non-ulcer dyspepsia (NUD) patients was used to show the model's performance in an actual application. The validity of models was evaluated using the 10-fold cross-validation. RESULTS: The presented model was superior to ordinary NB in all the simulation scenarios according to higher classification sensitivity and specificity in test data. The NB-BLCA model using Gibbs sampling accuracy was 87.77 (95% CI: 84.87-90.29). This index was estimated at 77.22 (95% CI: 73.64-80.53) and 74.71 (95% CI: 71.02-78.15) for the NB-BLCA model using the EM algorithm and ordinary NB classifier, respectively. CONCLUSIONS: When considering the modification of the NB classifier, incorporating a latent component into the model offers numerous advantages, particularly within medical and health-related contexts. By doing so, the researchers can bypass the extensive search algorithm and structure learning required in the local learning and structure extension approach. The inclusion of latent class variables allows for the integration of all attributes during model construction. Consequently, the NB-BLCA model serves as a suitable alternative to conventional NB classifiers when the assumption of independence is violated, especially in domains pertaining to health and medicine.

Asunto(s)

RESUMEN

More and more individuals are paying attention to the research on the emotional information found in micro-blog comments. TEXTCNN is growing rapidly in the short text space. However, because the training model of TEXTCNN model itself is not very extensible and interpretable, it is difficult to quantify and evaluate the relative importance of features and themselves. At the same time, word embedding can't solve the problem of polysemy at one time. This research suggests a microblog sentiment analysis method based on TEXTCNN and Bayes that addresses this flaw. First, the word embedding vector is obtained by word2vec tool, and based on the word vector, the ELMo word vector integrating contextual features and different semantic features is generated by ELMo model. Second, the local features of ELMo word vector are extracted from multiple angles by using the convolution layer and pooling layer of TEXTCNN model. Finally, the training task of emotion data classification is completed by combining Bayes classifier. On the Stanford Sentiment Classification Corpus data set SST (Stanford Sentiment Classification Corpus Data bank), the experimental findings demonstrate that the model in this paper is compared with TEXTCNN, LSTM, and LSTM-TEXTCNN models. The Accuracy, Precision, Recall, and F1-score of the experimental results of this research have all greatly increased. Their values are respectively 0.9813, 0.9821, 0.9804 and 0.9812, which are superior to other comparison models and can be effectively used for emotional accurate analysis and identification of events in microblog emotion analysis.

RESUMEN

Foraging as a natural visual search for multiple targets has increasingly been studied in humans in recent years. Here, we aimed to model the differences in foraging strategies between feature and conjunction foraging tasks found by Á. Kristjánsson et al. Bundesen proposed the theory of visual attention (TVA) as a computational model of attentional function that divides the selection process into filtering and pigeonholing. The theory describes a mechanism by which the strength of sensory evidence serves to categorise elements. We combined these ideas to train augmented Naïve Bayesian classifiers using data from Á. Kristjánsson et al. as input. Specifically, we attempted to answer whether it is possible to predict how frequently observers switch between different target types during consecutive selections (switches) during feature and conjunction foraging using Bayesian classifiers. We formulated 11 new parameters that represent key sensory and bias information that could be used for each selection during the foraging task and tested them with multiple Bayesian models. Separate Bayesian networks were trained on feature and conjunction foraging data, and parameters that had no impact on the model's predictability were pruned away. We report high accuracy for switch prediction in both tasks from the classifiers, although the model for conjunction foraging was more accurate. We also report our Bayesian parameters in terms of their theoretical associations with TVA parameters, πj (denoting the pertinence value), and ßi (denoting the decision-making bias).

Asunto(s)

RESUMEN

BACKGROUND AND OBJECTIVE: Diffusion tensor imaging (DTI) can be used to tract-wise map correlates of the sequential disease progression and, therefore, to assess disease stages of amyotrophic lateral sclerosis (ALS) in vivo. According to a threshold-based sequential scheme, a classification of ALS patients into disease stages is possible, however, several patients cannot be staged for methodological reasons. This study aims to implement a multivariate Bayesian classification algorithm for disease stage prediction at an individual ALS patient level based on DTI metrics of involved tract systems to improve disease stage mapping. METHODS: The analysis of fiber tracts involved in each stage of ALS was performed in 325 ALS patients and 130 age- and gender-matched healthy controls. Based on Bayes' theorem and in accordance with the sequential disease progression, a multistage classifier was implemented. Patients were categorized into in vivo DTI stages using the threshold-based method and the Bayesian algorithm. By the margin of confidence, the reliability of the Bayesian categorizations was accessible. RESULTS: Based on the Bayesian multistage classifier, 88% of all ALS patients could be assigned into an ALS stage compared to 77% using the threshold-based staging scheme. Additionally, the confidence of all classifications could be estimated. CONCLUSIONS: By the application of the multi-stage Bayesian classifier, an individualized in vivo cerebral staging of ALS patients was possible based on the sequentially involved tract systems and, furthermore, the reliability of the respective classifications could be determined. The Bayesian classification algorithm is an improvement of the threshold-based staging method and could provide a framework for extending the DTI-based in vivo cerebral staging in ALS.

Asunto(s)

RESUMEN

Healthcare AI systems exclusively employ classification models for disease detection. However, with the recent research advances into this arena, it has been observed that single classification models have achieved limited accuracy in some cases. Employing fusion of multiple classifiers outputs into a single classification framework has been instrumental in achieving greater accuracy and performing automated big data analysis. The article proposes a bit fusion ensemble algorithm that minimizes the classification error rate and has been tested on various datasets. Five diversified base classifiers k- nearest neighbor (KNN), Support Vector Machine (SVM), Multi-Layer Perceptron (MLP), Decision Tree (D.T.), and Naïve Bayesian Classifier (N.B.), are used in the implementation model. Bit fusion algorithm works on the individual input from the classifiers. Decision vectors of the base classifier are weighted transformed into binary bits by comparing with high-reliability threshold parameters. The output of each base classifier is considered as soft class vectors (CV). These vectors are weighted, transformed and compared with a high threshold value of initialized δ = 0.9 for reliability. Binary patterns are extracted, and the model is trained and tested again. The standard fusion approach and proposed bit fusion algorithm have been compared by average error rate. The error rate of the Bit-fusion algorithm has been observed with the values 5.97, 12.6, 4.64, 0, 0, 27.28 for Leukemia, Breast cancer, Lung Cancer, Hepatitis, Lymphoma, Embryonal Tumors, respectively. The model is trained and tested over datasets from UCI, UEA, and UCR repositories as well which also have shown reduction in the error rates.

Asunto(s)

RESUMEN

Pulmonary tuberculosis (TB) is one of the top 10 causes of death worldwide caused by an infection. TB is curable with an adequate diagnosis, normally performed through bacilloscopies. Automate TB diagnosis implies bacilli detection and counting usually based on smear images processing and artificial intelligence. Works reported in the literature usually consider images with similar coloring characteristics, which are difficult to obtain due to the Ziehl - Neelsen staining method variations (excess or deficiency of coloration), provoking errors in the bacilli segmentation. This paper presents an image preprocessing technique, based on simple, fast and well-known processing techniques, to improve and standardize the contrast in the Acid-Fast Bacilli (AFB) images used to diagnose TB; these techniques are used previously to the segmentation stage to obtain accurate results. The results are validated with and without the preprocessing stage by the Jaccard index, pixel detection accuracy and UAC obtained in an Artificial Neural Network (ANN) and a Bayesian classifier with Gaussian mixture model (GMM). Obtained results indicate that the proposed approach can be applied to automate the Tuberculosis diagnostic.

Asunto(s)

RESUMEN

Aeration system is the main energy consumer in a wastewater treatment process. In this paper, the Naive Bayes classification (NBC) algorithm and response surface method (RSM) were firstly used to establish a methodology to improve the aeration efficiency and estimate effluent quality. Lab-scale experiments were conducted to verify the model. The errors between experimental values and predicted values were 3.36, -0.67 and -3.78% at operating temperatures of 20, 30 and 35 °C, indicating the applicability. To further elucidate the biological mechanisms of the experimental results, the microbial community composition was investigated under various operating conditions, the results shows that aerobic heterotrophic bacteria (HET) activity and COD removal efficiency were promoted at 30 °C. AOB and NOB activity and NH4+-N removal efficiency were promoted at 30-35 °C. These findings together suggest that operating temperature is crucial for activated sludge treatment, which should be considered when regulating DO content or aeration rate in practical application.

Asunto(s)

RESUMEN

Introduction: Phosphoinositide 3-kinase gamma (PI3Kγ) has been regarded as a promising drug target for the treatment of various diseases, and the diverse physiological roles of class I PI3K isoforms (α, ß, δ, and γ) highlight the importance of isoform selectivity in the development of PI3Kγ inhibitors. However, the high structural conservation among the PI3K family makes it a big challenge to develop selective PI3Kγ inhibitors. Objectives: A novel machine learning-based virtual screening with multiple PI3Kγ protein structures was developed to discover novel PI3Kγ inhibitors. Methods: A large chemical database was screened using the virtual screening model, the top-ranked compounds were then subjected to a series of bio-evaluations, which led to the discovery of JN-KI3. The selective inhibition mechanism of JN-KI3 against PI3Kγ was uncovered by a theoretical study. Results: 49 hits were identified through virtual screening, and the cell-free enzymatic studies found that JN-KI3 selectively inhibited PI3Kγ at a concentration as low as 3,873 nM but had no inhibitory effect on Class IA PI3Ks, leading to the selective cytotoxicity on hematologic cancer cells. Meanwhile, JN-KI3 potently blocked the PI3K signaling, finally led to distinct apoptosis of hematologic cell lines at a low concentration. Lastly, the key residues of PI3Kγ and the structural characteristics of JN-KI3, which both would influence γ isoform-selective inhibition, were highlighted by systematic theoretical studies. Conclusion: The developed virtual screening model strongly manifests the robustness to find novel PI3Kγ inhibitors. JN-KI3 displays a specific cytotoxicity on hematologic tumor cells, and significantly promotes apoptosis associated with the inhibition of the PI3K signaling, which depicts PI3Kγ as a potential target for the hematologic tumor therapy. The theoretical results reveal that those key residues interacting with JN-KI3 are less common compared to most of the reported PI3Kγ inhibitors, indicating that JN-KI3 has novel structural characteristics as a selective PIK3γ inhibitor.

Asunto(s)

RESUMEN

Worldwide honeybees (Apis mellifera L.) are one of the most widely kept domesticated animals, supporting domestic and commercial livelihoods through the production of honey and wax, as well as in the delivery of pollination services to crops. Quantifying which plant species are foraged upon by honeybees provides insights into their nutritional status as well as patterns of landscape scale habitat utilization. Here we outline a rapid and reproducible methodology for identifying environmental DNA (eDNA) originating principally from pollen grains suspended within honey. The process is based on a DNA extraction incorporating vacuum filtration prior to universal eukaryotic internal transcribed spacer 2 region (ITS2) amplicon generation, sequencing and identification. To provide a pre-cursor to sequence phylotyping, we outline systems for error-corrected processing amplicon sequence variant abundance tables that removes chimeras. This methodology underpins the new UK National Honey Monitoring Scheme.•We compare the efficacy and speed of centrifugation and filtration systems for removing pollen from honey samples as a precursor to plant DNA barcoding.•We introduce the 'HONEYPI' informatics pipeline, an open access resource implemented in python 2.7, to ensure long-term reproducibility during the process of amplicon sequence variant classification.

RESUMEN

MALDI-TOF MS is one of the major methods for clinical fungal identification, but it is currently only suitable for pure cultures of isolated strains. However, multiple fungal coinfections might occur in clinical practice. Some fungi involved in coinfection, such as Candida krusei and Candida auris, are intrinsically resistant to certain drugs. Identifying intrinsically resistant fungi from coinfected mixed cultures is extremely important for clinical treatment because different treatment options would be pursued accordingly. In this study, we counted the peaks of various species generated by Bruker Daltonik MALDI Biotyper software and accordingly constructed a modified naïve Bayesian classifier to analyze the presence of C. krusei and C. auris in simulated mixed samples. When reasonable parameters were fixed, the modified naïve Bayesian classifier effectively identified C. krusei and C. auris in the mixed samples (sensitivity 93.52%, specificity 92.5%). Our method not only provides a viable solution for identifying the two highlighted intrinsically resistant Candida species but also provides a case for the use of MALDI-TOF MS for analyzing coinfections of other species.

Asunto(s)

RESUMEN

The multimedia service company, Netflix, increased the number of new subscribers during the Coronavirus pandemic age. Intrusion detection systems for multimedia platforms can prevent the platform from network attacks. An intelligent intrusion detection system is proposed for the security IP Multimedia Subsystem (IMS) based on machine learning technology. For increasing the accuracy of the classifiers, it is vital to select the critical features to construct the intrusion detection system. Two-class classifiers, including the Decision Tree, Support Vector Machine, and Naive Bayesian, are selected to evaluate intrusion detection accuracy. According to the three classifiers' accuracy values, the most critical features are selected based on the features' ranking orders. Six critical features are selected:Service, dst_host_same_srv_rate, Flag, Protocol Type, Dst_host_rerror_rate, and Count. Numerical comparison with state_of_the_art shows that critical features improve intrusion detection accuracy, which can be better than the deep learning method.

RESUMEN

Selection of toxicity endpoints affects outcomes of risk assessment. Scientific decisions based on more holistic evidence is preferable for designing bioassay batteries rather than subjective selections, particularly when systems are poorly understood. Here, we propose a novel event-driven taxonomy (EDT)-based text mining tool to prioritize stressors likely to elicit water quality deterioration. The tool integrated automated literature collection, natural language processing using adverse outcome pathway-based toxicological terminologies and machine learning to classify event drivers (EDs). From aquatic toxicity assessments within China over the past decade, we gathered over 14 000 sources of information. With a dictionary that included 1039 toxicological terms, 15 bioassay-related modes of actions were mapped, yet less than half of the bioassays could be elucidated by available adverse outcome pathways. To fill these mechanistic knowledge gaps, we developed a Naïve Bayesian ED-classifier to annotate apical responses. The classifier's 4-fold cross-validation reached 74% accuracy and labeled 85% bioassays as 26 EDs. Narcosis, estrogen receptor-, and aryl hydrogen receptor-mediators were the major EDs in aquatic systems across China, whereas individual regions had distinct ED fingerprints. The EDT-based tool provides a promising diagnostic strategy to inform region-specific bioassay design and selection for water quality assessments in a big data era.

Asunto(s)

RESUMEN

We develop a predictive prognosis model to support medical experts in their clinical decision-making process in Intensive Care Units (ICUs) (a) to enhance early mortality prediction, (b) to make more efficient medical decisions about patients at higher risk, and (c) to evaluate the effectiveness of new treatments or detect changes in clinical practice. It is a machine learning hierarchical model based on Bayesian classifiers built from some recorded features of a real-world ICU cohort, to bring about the assessment of the risk of mortality, also predicting destination at ICU discharge if the patient survives, or the cause of death otherwise, constructed as an ensemble of five base Bayesian classifiers by using the average ensemble criterion with weights, and we name it the Ensemble Weighted Average (EWA). We compare EWA against other state-of-the-art machine learning predictive models. Our results show that EWA outperforms its competitors, presenting in addition the advantage over the ensemble using the majority vote criterion of allowing to associate a confidence level to the provided predictions. We also prove the convenience of locally recalibrate from data the standard model used to predict the mortality risk based on the APACHE II score, although as a predictive model it is weaker than the other.

Asunto(s)

RESUMEN

PURPOSE: New radiation therapy protocols, in particular adaptive, focal or boost brachytherapy treatments, require determining precisely the position and orientation of the implanted radioactive seeds from real-time ultrasound (US) images. This is necessary to compare them to the planned one and to adjust automatically the dosimetric plan accordingly for next seeds implantations. The image modality, the small size of the seeds, and the artifacts they produce make it a very challenging problem. The objective of the presented work is to setup and to evaluate a robust and automatic method for seed localization in three-dimensional (3D) US images. METHODS: The presented method is based on a prelocalization of the needles through which the seeds are injected in the prostate. This prelocalization allows focusing the search on a region of interest (ROI) around the needle tip. Seeds localization starts by binarizing the ROI and removing false positives using, respectively, a Bayesian classifier and a support vector machine (SVM). This is followed by a registration stage using first an iterative closest point (ICP) for localizing the connected set of seeds (named strand) inserted through a needle, and secondly refining each seed position using sum of squared differences (SSD) as a similarity criterion. ICP registers a geometric model of the strand to the candidate voxels while SSD compares an appearance model of a single seed to a subset of the image. The method was evaluated both for 3D images of an Agar-agar phantom and a dataset of clinical 3D images. It was tested on stranded and on loose seeds. RESULTS: Results on phantom and clinical images were compared with a manual localization giving mean errors of 1.09 ± 0.61 mm on phantom image and 1.44 ± 0.45 mm on clinical images. On clinical images, the mean errors of individual seeds orientation was 4.33 ± 8 . 51 ∘ . CONCLUSIONS: The proposed algorithm for radioactive seed localization is robust, tested on different US images, accurate, giving small mean error values, and returns the five cylindrical seeds degrees of freedom.

Asunto(s)

RESUMEN

DNA barcodes are standardized sequences that range between 400 and 800 bp, vary at different taxonomic levels, and make it possible to assign sequences to species that have been previously taxonomically characterized. Several DNA barcodes have been postulated for plants, nonetheless, their classification potential has not been evaluated for metabarcoding, and as a result, it would appear as none of them excels above the others in this area. One tool that has been widely used and served as a baseline when evaluating new approaches is Naïve Bayesian Classifiers (NBC). The present study aims at evaluating the classification power of several plant chloroplast genetic markers that have been proposed as barcodes (trnL, rpoB, rbcL, matK, psbA-trnH, and psbK) using an NBC. We performed the classification at different taxonomic levels, and identified problematic genera when resolution was desired. We propose matK and trnL as potential candidate markers with resolution up to genus level. Some problematic genera within certain families could lead to the misclassification no matter which marker is used (i.e., Aegilops, Gueldenstaedtia, Helianthus, Oryza, Shorea, Thysananthus, and Triticum). Finally, we suggest recommendations for the taxonomic identification of plants in samples with potential mixtures.

RESUMEN

Quantitative characterization of disease progression using longitudinal data can provide long-term predictions for the pathological stages of individuals. This work studies the robust modeling of Alzheimer's disease progression using parametric methods. The proposed method linearly maps the individual's age to a disease progression score (DPS) and jointly fits constrained generalized logistic functions to the longitudinal dynamics of biomarkers as functions of the DPS using M-estimation. Robustness of the estimates is quantified using bootstrapping via Monte Carlo resampling, and the estimated inflection points of the fitted functions are used to temporally order the modeled biomarkers in the disease course. Kernel density estimation is applied to the obtained DPSs for clinical status classification using a Bayesian classifier. Different M-estimators and logistic functions, including a novel type proposed in this study, called modified Stannard, are evaluated on the data from the Alzheimer's Disease Neuroimaging Initiative (ADNI) for robust modeling of volumetric magnetic resonance imaging (MRI) and positron emission tomography (PET) biomarkers, cerebrospinal fluid (CSF) measurements, as well as cognitive tests. The results show that the modified Stannard function fitted using the logistic loss achieves the best modeling performance with an average normalized mean absolute error (NMAE) of 0.991 across all biomarkers and bootstraps. Applied to the ADNI test set, this model achieves a multiclass area under the ROC curve (AUC) of 0.934 in clinical status classification. The obtained results for the proposed model outperform almost all state-of-the-art results in predicting biomarker values and classifying clinical status. Finally, the experiments show that the proposed model, trained using abundant ADNI data, generalizes well to data from the National Alzheimer's Coordinating Center (NACC) with an average NMAE of 1.182 and a multiclass AUC of 0.929.

Asunto(s)

RESUMEN

PURPOSE: Focal cortical dysplasia (FCD) is a malformation of cortical development that often causes pharmacologically intractable epilepsy. However, FCD lesions are frequently characterized by minor structural abnormalities that can easily go unrecognized, making diagnosis difficult. Therefore, many epileptic patients have had pathologically confirmed FCD lesions that appeared normal in pre-surgical fluid-attenuated inversion recovery (FLAIR) magnetic resonance (MR) studies. Such lesions are called "FLAIR-negative." This study aimed to improve the detection of histopathologically verified FCD in a sample of patients without visually appreciable lesions. METHODS: The technique first extracts a series of features from a FLAIR image. Then, three naive Bayesian classifiers with probability (NBCP) are trained based on different numbers of feature maps to classify voxels as lesional or healthy voxels and assign the lesions a probability of correct classification. This method classifies the three-dimensional (3D) images of all patients using leave-one-out cross-validation (LOOCV). Finally, the 3D lesion probability map, including epileptogenic lesions, is obtained by removing false-positive voxel outliers using the morphological method. The performance of the NBCP was assessed for quantitative analysis by specificity, accuracy, recall, precision, and Dice coefficient in subject-wise, lesion-wise, and voxel-wise manners. RESULTS: The best detection results were obtained by using four features: cortical thickness, symmetry, K-means, and modified texture energy. There were eight lesions in seven patients. The subject-wise sensitivity of the proposed method was 85.71% (6/7). Seven out of eight lesions were detected, so the lesion-wise sensitivity was 87.50% (7/8). No significant differences in effectiveness were found between automated lesion detection using four features and lesion detection using manual segmentation, as voxels were quantitatively analyzed in terms of specificity (mean ± SD = 99.64 ± 0.13), accuracy (mean ± SD = 99.62 ± 0.14), recall (mean ± SD = 73.27 ± 26.11), precision (mean ± SD = 11.93 ± 8.16), and Dice coefficient (mean ± SD = 22.82 ± 15.57). CONCLUSION: We developed a novel automatic voxel-based method to improve the detection of FCD FLAIR-negative lesions. To the best of our knowledge, this study is the first to detect FCD lesions that appear normal in pre-surgical 3D high-resolution FLAIR images alone with a limited number of radiomics features. We optimized the algorithm and selected the best prior probability to improve the detection. For non-temporal lobe epilepsy (non-TLE) patients, lesions could be accurately located, although there were still false-positive areas.

Asunto(s)