Discriminant Model Research Articles

Pix2Pix GAN for Realistic Lung CT scans Generation from COVID-19 Lesion Masks

GANs have recently become one of the most promising directions of machine learning for generative tasks in deep learning. By using architectures such as CNN, GANs can learn relationships in input data without any human intervention and generate new examples that are similar to the original data set. This paper aims at understanding what GANs are and how they work, and the different types of GAN models with a special focus on their application in image synthesis, style transfer, and text-to-image synthesis. The GAN framework consists of two neural networks: The two primary components of the model are the Generator and the Discriminator. The generator’s goal is to generate ‘realistic’ data such that the discriminator cannot differentiate between the two, while the discriminator’s goal is to distinguish between the real and the fake data. Some other GAN variants include Vanilla GAN, conditional GAN (CGAN), Deep Convolutional GAN (DCGAN), Laplacian Pyramid GAN (LAPGAN) and Super Resolution GAN (SRGAN), all of which have made significant improvements in producing convincing images. The structures of the generator model, which produces data from noise, and the discriminator model, which checks the validity of the data produced, are also considered. The study focuses on the concept of loss functions including the generator loss, discriminator loss and minimax loss in the training of GANs. Pre- processing the data for GAN training is a very crucial step that requires careful consideration as well as data cleaning. This is because when using content and adversarial loss functions simultaneously, it is easier to find the right balance between the photorealism of the images and their structural similarity to the source material. By training carefully and iterating over the process, GANs exhibit high performance in sample generation, thereby leading to the new state of generative modeling. It will be beneficial for any researcher or practitioner by providing an understanding of artificial intelligence and generative modeling

Risk factors and discrimination model for screen exposure among children under 3 years: findings from Fujian province in China.

Compliance with screen time guidelines among children worldwide remains low, and there is insufficient evidence on the current prevalence in China. This study aimed to investigate the prevalence of compliance with screen time guidelines among children under 3 years old in Fujian Province, East China, identify risk factors and their independent effects, and develop a risk discrimination model for targeted interventions. A cross-sectional survey was conducted among low-income families recruited from welfare programs at 96 sites in both urban and rural areas of Fujian Province, China. Face-to-face interviews gathered sociodemographic data, lifestyle information, attitudes towards screen exposure, and details on screen media device ownership. A multivariable logistic regression model was employed to identify independent risk factors for compliance with screen time guidelines, while the area under the receiver operating characteristic curve (AUC) was used to evaluate the model's discrimination ability. A total of 4,707 children participated in the survey. The rates of compliance with screen time guidelines were 56.8% for children under 1 year old, 18.8% for those between 1 and 2 years old, and 81.9% for those between 2 and 3 years old. The multivariable regression analysis identified negative attitudes towards screen exposure, co-viewing and engagement, as well as single child, as significant positive independent factors for compliance with the guidelines. The risk discrimination model demonstrated good performance, with an AUC of 0.845 and 0.812 in the two younger age groups, but showed medium discrimination with an AUC of 0.691 for children between 2 and 3 years old. Compliance with screen time guidelines among young children in Fujian Province, East China, is generally adequate, but notably low among children between 1 and 2 years old. Targeted interventions are needed to improve compliance, particularly for this age group.

GeM-LR: Discovering predictive biomarkers for small datasets in vaccine studies.

Despite significant progress in vaccine research, the level of protection provided by vaccination can vary significantly across individuals. As a result, understanding immunologic variation across individuals in response to vaccination is important for developing next-generation efficacious vaccines. Accurate outcome prediction and identification of predictive biomarkers would represent a significant step towards this goal. Moreover, in early phase vaccine clinical trials, small datasets are prevalent, raising the need and challenge of building a robust and explainable prediction model that can reveal heterogeneity in small datasets. We propose a new model named Generative Mixture of Logistic Regression (GeM-LR), which combines characteristics of both a generative and a discriminative model. In addition, we propose a set of model selection strategies to enhance the robustness and interpretability of the model. GeM-LR extends a linear classifier to a non-linear classifier without losing interpretability and empowers the notion of predictive clustering for characterizing data heterogeneity in connection with the outcome variable. We demonstrate the strengths and utility of GeM-LR by applying it to data from several studies. GeM-LR achieves better prediction results than other popular methods while providing interpretations at different levels.

Research on Key Technologies of Image Steganography Based on Simultaneous Deception of Vision and Deep Learning Models

As machine learning continues to evolve, traditional image steganography techniques find themselves increasingly unable to meet the dual challenge of deceiving both the human visual system and machine learning models. In response, this paper introduces the Visually Robust Image Steganography (VRIS) model, specifically tailored for this dual deception task. The VRIS model elevates visual security through the use of specialized feature extraction and processing methodologies. It meticulously conducts feature-level fusion between secret images and cover images, ensuring a high level of visual similarity between them. To effectively mislead machine learning models, the VRIS model incorporates a sophisticated strategy utilizing random noise factors and discriminators. This involves adding controlled amounts of random Gaussian noise to the encrypted image, thereby enhancing the difficulty for machine learning frameworks to recognize it. Furthermore, the discriminator is trained to discern between the noise-infused encrypted image and the original cover image. Through adversarial training, the discriminator and VRIS model refine each other, successfully deceiving the machine learning systems. Additionally, the VRIS model presents an innovative method for extracting and reconstructing secret images. This approach safeguards secret information from unauthorized access while enabling legitimate users to non-destructively extract and reconstruct images by leveraging multi-scale features from the encrypted image, combined with advanced feature fusion and reconstruction techniques. Experimental results validate the effectiveness of the VRIS model, achieving high PSNR and SSIM scores on the LFW dataset and demonstrating significant deception capabilities against the ResNet50 model on the Mini-ImageNet dataset, with an impressive misclassification rate of 99.24%.

Investigating the effect of non-resonant background variation on the CARS data analysis of bacteria samples and classification using machine learning

Non-resonant background (NRB) plays a significant role in coherent anti-Stokes Raman scattering (CARS) spectroscopic applications. All the recent works primarily focused on removing the NRB using different deep learning methods, and only one study explored the effect of NRB. Hence, in this work, we systematically investigated the impact of NRB variation on Raman signal retrieval. The NRB is simulated as a linear function with different strengths relative to the resonant Raman signal, and the variance also changes for each NRB strength. The resonant part of nonlinear susceptibility is extracted from real experimental Raman data; hence, the simulated CARS data better approximate the experimental CARS spectra. Then, the corresponding Raman signal is retrieved by four different methods: maximum entropy method (MEM), Kramers-Kronig (KK), convolutional neural network (CNN), and long short-term memory (LSTM) network. Pearson correlation measurements and principal component analysis combined with linear discriminant analysis modeling revealed that MEM and KK methods have an edge over LSTM and CNN for higher NRB strengths. It is also demonstrated that normalizing the input data favors LSTM and CNN predictions. In contrast, background removal from the predictions significantly influenced Pearson correlation but not the classification accuracies for MEM and KK. Further, the LSTM performance is found to be limited and can only be applied for low NRB strengths. This comprehensive study has the potential to impact the CARS spectroscopy and microscopy applications in different areas.

Abstract 4137726: A Deep Learning Digital Biomarker for Mitral Valve Prolapse using Echocardiogram Videos

Background: Mitral valve prolapse (MVP) has a prevalence of 2–3% and is a risk factor for heart failure and sudden death, but MVP diagnosis by transthoracic echocardiography (TTE) requires time and clinical expertise. We trained a deep learning model to classify MVP from TTE videos. Methods: DROID-MVP is a convolutional neural network trained to classify MVP using echocardiographer labels using 973,531 digital videos (parasternal long axis and apical views) from 45,657 studies performed in 15,728 cardiology patients at Massachusetts General Hospital. We validated DROID-MVP in 1,726 cardiology patients (4,869 studies) and 8,903 primary care patients (8,903 studies), and tested associations between predicted MVP score (range 0-1) and mitral regurgitation (MR) severity and left atrial (LA) anterior-posterior diameter in primary care patients with available measurements. Results: Of 15,728 patients (6,029 [38%] women; mean age at first TTE 61 ± 17 years) in the training set, 729 (4.6%) had at least 1 study with MVP. DROID-MVP identified MVP in both the cardiology (area under the receiver operating characteristic curve [AUC] 0.955 [95% confidence interval: 0.939-0.970]; average precision [AP] 0.716 [0.649-0.776]; prevalence 0.035) and primary care (AUC 0.966 [0.955-0.978]; AP 0.668 [0.601-0.730]; prevalence 0.022) test sets ( Figure 1 ). Discrimination persisted across strata of MR severity (AUC range 0.877-0.987). High (>0.67) vs low (<0.33) MVP score was associated with higher odds of moderate/severe MR (odds ratio 20.1 [12.2-33.4], p<0.001) and higher LA diameter (2.7 [0.9-4.4] mm, p<0.001; Figure 2 ), after adjusting for age, sex, height, and weight. Conclusions: A deep learning model identifies MVP from TTE videos with excellent discrimination and higher model predictions are associated with greater MR severity and LA size. Our work demonstrates how deep learning can automate MVP diagnosis and may serve as a digital biomarker of MVP-associated structural abnormalities.

Assessing groundwater quality in the northwestern coastal area of Sri Lanka using multivariate statistical techniques

The objective of the present study was to assess groundwater quality in the northwestern coastal region of Sri Lanka by analyzing hydrogeochemical data using multivariate techniques. The study examined the relationship between water quality and sources influencing groundwater quality and proposed useful statistical tools for water resource management. Descriptive, comparative, correlation, factor analysis (FA), and discriminant analysis (DA) were among the techniques used. The water quality in terms of electrical conductivity (EC), total dissolved solids (TDS), and dissolved nitrate and phosphate concentrations of the northwestern coastal area, where different aquifer conditions prevailed, were compared, and differences in the mean concentrations of the above parameters were identified. The study also employed ridge regression and support vector machines (SVM) to predict nitrate levels and validate the results using a 70:30 train/test split. The analysis found that the mean concentrations of EC, TDS, and HCO3- are higher in the sandy aquifer area, while Fe2+ and PO43- are higher in the regolith aquifer area. EC, TDS, PO43-, and HCO3- in the hard-rock aquifer area exceed permissible limits. Factor analysis identified four factors contributing to pollution, and a linear discriminant model classified all observations with 99.9% accuracy. The water quality index for Kalpitiya, Vanathavillu, and Katana-Negombo areas was 68.1, 99.2, and 287.9, respectively, while the World Health Organization (WHO) cutoff was 50. The use of multivariate statistical techniques demonstrated the value of analyzing hydrogeochemical data to gain valuable insights into the complex interplay of factors affecting groundwater quality. The SVM model, using a linear kernel, was identified as the best model to predict nitrate levels. A combination of statistical analyses reveals that, with the exclusion of a few wells, the groundwater in the study area is polluted and inappropriate for human consumption or culinary purposes. Nonetheless, the inhabitants persist in utilizing it as their primary source of drinking water. This study reveals the combination of statistical analyses is useful for the management of groundwater resources.

Development of FTIR-ATR spectra and PLS regression combination model for discrimination of pure and adulterated acacia honey

A Fourier Transform Infrared Spectroscopy and attenuated total reflectance (FTIR-ATR)-based chemometric model was evaluated for the rapid identification and estimation of adulterants in honey. Two types of honey, bee honey and stingless bee honey, were adulterated with acid adulterants (acetic acid, citric acid, and tamarind extract) at different concentrations of 1%, 3%, 5%, and 7% and sugar adulterants (liquid corn syrup, cane sugar, palm sugar, and inverted sugar) at different concentrations of 1%, 2%, and 3%. The physicochemical properties (Brix, moisture content, pH, and free acidity) and sugar contents (glucose, fructose, and sucrose) of each type of honey and their adulterated samples were determined before FTIR analysis. For all the samples, FTIR spectra were acquired in the mid-infrared range (400-3600 cm−1) and the spectra obtained were subjected to partial least squares (PLS) analysis to develop the model for the determination of adulterants in honey samples. The PLS model produced high coefficient of determination (R2) for bee honey adulterated with acetic acid (0.95), citric acid (0.98), tamarind extract (0.97), and stingless bee honey adulterated with liquid corn syrup (0.99), inverted sugar (0.91), cane sugar (0.86), and palm sugar (0.77). A recognition model was developed for mixture detection and verified through the physicochemical analysis. The combination of FTIR-ATR spectra and the PLS regression model can be a fast, reliable, nondestructive method to detect honey adulteration.

Nonlinear Dynamic Process Monitoring Based on Discriminative Denoising Autoencoder and Canonical Variate Analysis

Modern industrial processes are characterized by increasing complexity, often exhibiting pronounced dynamic behaviors and significant nonlinearity. Addressing these dynamic and nonlinear characteristics is essential for effective process monitoring. However, many existing methods for process monitoring and fault diagnosis are insufficient in handling these challenges. In this article, we present a novel process monitoring approach, CVA-DisDAE, which integrates an improved Denoising Autoencoder (DAE) with Canonical Variate Analysis (CVA) to address the challenges posed by dynamic behaviors and nonlinear relationships in industrial processes. First, CVA is employed to reduce data dimensionality and minimize information redundancy by maximizing correlations between past and future observations, thereby effectively capturing process dynamics. Following this, we introduce a discriminative DAE model (DisDAE) designed to serve as a semi-supervised denoising autoencoder for precise feature extraction. This is achieved by incorporating both between-class separability and within-class variability into the traditional DAE framework. The key distinction between the proposed DisDAE and the conventional DAE lies in the integration of a linear discriminant analysis (LDA) penalty into the DAE’s loss function, resulting in extracted features that are more conducive to fault classification. Finally, we validate the effectiveness of the proposed semi-supervised dynamic process monitoring approach through its application to the Tennessee Eastman benchmark process, demonstrating its superior performance.

Rapid identification of fragrant rice using starch flavor compound via NIR spectroscopy coupled with GC–MS and Badh2 genotyping

The identification of fragrant rice varieties using near-infrared reflectance spectroscopy (NIRS) models has attracted extensive attention from regulatory authorities worldwide. In this study, 138 fragrant and 54 nonfragrant rice varieties were planted in the same region and distinguished using sensory evaluation, gas chromatography–mass spectrometry analysis, and betaine aldehyde dehydrogenase 2 (Badh2) genotyping. Then, the 2-acetyl−1-pyrroline (2-AP) content was assessed based on partial least-squares discriminant (PLS-DA) models generated after 2nd individually or combined with SNV/MSC/smoothing preprocessing successfully classified fragrant rice both in the calibration and predictive sets. Moreover, design of experiments (DoE)-based preprocessing selection was employed as an effective strategy to optimize the calibration models compared with the one variable at a time (OVAT) method. Further, Badh2 genotype sample screening assisted with identifying authentic fragrant rice and guaranteed the NIRS model's prediction accuracy in identifying fragrant rice. In conclusion, the high throughput PLS-DA multivariate method coupled with NIRS data was applied to identify fragrant rice varieties in routine monitoring and was effective, accurate, and rapid.

Teaching competencies of automotive engineering vocational school teachers in East Kalimantan

This study aimed to examine the factors that classify teacher competencies, based on good and bad educator certificates. The method used in this research is discriminant analysis. This study used a questionnaire survey method. The sample in this study was 54 teachers of the Light Automotive Engineering Vocational High School. The findings of this study revealed several key points: first, there is a significant average difference between competent and non-competent teachers based on the factors evaluated. Second, a discriminant function model was developed from the data, which effectively distinguishes between teacher competencies. Lastly, the model demonstrated a high prediction accuracy, successfully classifying teachers as competent or non-competent with a high degree of reliability. Therefore, this discriminant model can be utilized to assess teacher competency effectively. Keywords: Automotive teacher competence; competencies; teachers; vocational high schools.

Comparative Blood Profiling Based on ATR-FTIR Spectroscopy and Chemometrics for Differential Diagnosis of Patients with Amyotrophic Lateral Sclerosis—Pilot Study

Amyotrophic lateral sclerosis (ALS) is a motor neurodegenerative disease characterized by poor prognosis. Currently, screening and diagnostic methods for ALS remain challenging, often leading to diagnosis at an advanced stage of the disease. This delay hinders the timely initiation of therapy, negatively impacting patient well-being. Additionally, misdiagnosis with other neurodegenerative disorders that present similar profiles often occurs. Therefore, there is an urgent need for a cost-effective, rapid, and user-friendly tool capable of predicting ALS onset. In this pilot study, we demonstrate that infrared spectroscopy, coupled with chemometric analysis, can effectively identify and predict disease profiles from blood samples drawn from ALS patients. The selected predictive spectral markers, which are used in various discriminant models, achieved an AUROC sensitivity of almost 80% for distinguishing ALS patients from controls. Furthermore, the differentiation of ALS at both the initial and advanced stages from other neurodegenerative disorders showed even higher AUROC values, with sensitivities of 87% (AUROC: 0.70–0.97). These findings highlight the elevated potential of ATR-FTIR spectroscopy for routine clinical screening and early diagnosis of ALS.

HRMAS-NMR-Based Metabolomics Approach to Discover Key Differences in Cow and Goat Milk Yoghurt Metabolomes.

This study highlights the differences in the metabolomes of cow milk yoghurt (CY) and goat milk yoghurt (GY) using a nuclear magnetic resonance (NMR)-based metabolomic approach. The 1H HRMAS-NMR spectrum displayed 21 metabolites comprising organic acids, sugars, amino acids, amino acid derivatives and phospholipids. The orthogonal partial least squares discriminant analysis model clearly separated CY and GY groups, implying differences in metabolite composition. The corresponding Variable Importance in Projection (VIP) plot revealed that choline, sn-glycero-3-phosphocholine, O-phosphocholine, fucose, citrate, sucrose, glucose and lactose mainly contributed to the group separation (VIP > 1). Hierarchical cluster analysis further confirmed the metabolome similarities and differences between CY and GY. Additionally, 12 significantly differential metabolites (with a fold change > 1.5 and p-value < 0.05) were identified, with 1 downregulated and 11 upregulated. Pathway impact analysis revealed the correlation of significant metabolites with starch and sucrose metabolism, galactose metabolism, and the citrate cycle. Furthermore, receiver operating characteristic curve analysis identified eight metabolites (choline, sn-glycero-3-phosphocholine, fucose, O-phosphocholine, glucose, citrate, 2-oxoglutarate, lactose and sucrose) as candidate biomarkers. This study represents the first utilization of HRMAS-NMR to analyze the metabolomic profiles of yoghurt made from cow and goat milk. In conclusion, these findings provide preliminary information on how NMR-based metabolomics can discriminate the metabolomes of yoghurt prepared from the milk of two different animals, which may be valuable for authenticity and adulteration assessments.

CLAP. I. Resolving miscalibration for deep learning-based galaxy photometric redshift estimation

Obtaining well-calibrated photometric redshift probability densities for galaxies without a spectroscopic measurement remains a challenge. Deep learning discriminative models, typically fed with multi-band galaxy images, can produce outputs that mimic probability densities and achieve state-of-the-art accuracy. However, several previous studies have found that such models may be affected by miscalibration, an issue that would result in discrepancies between the model outputs and the actual distributions of true redshifts. Our work develops a novel method called the C ontrastive L earning and A daptive KNN for P hotometric Redshift (CLAP) that resolves this issue. It leverages supervised contrastive learning (SCL) and $k$-nearest neighbours (KNN) to construct and calibrate raw probability density estimates, and implements a refitting procedure to resume end-to-end discriminative models ready to produce final estimates for large-scale imaging data, bypassing the intensive computation required for KNN. The harmonic mean is adopted to combine an ensemble of estimates from multiple realisations for improving accuracy. Our experiments demonstrate that CLAP takes advantage of both deep learning and KNN, outperforming benchmark methods on the calibration of probability density estimates and retaining high accuracy and computational efficiency. With reference to CLAP, a deeper investigation on miscalibration for conventional deep learning is presented. We point out that miscalibration is particularly sensitive to the method-induced excessive correlations among data instances in addition to the unaccounted-for epistemic uncertainties. Reducing the uncertainties may not guarantee the removal of miscalibration due to the presence of such excessive correlations, yet this is a problem for conventional methods rather than CLAP. These discussions underscore the robustness of CLAP for obtaining photometric redshift probability densities required by astrophysical and cosmological applications. This is the first paper in our series on CLAP.

Discrimination Model Construction for Non-Lactational Mastitis and Breast Cancer Based on Imaging Features.

Aims/Background The clinical presentation of non-lactational mastitis (NLM) shares similarities with some symptoms and examination results of breast cancer (BC), which can lead to misdiagnosis or delayed treatment. Current studies on breast lesions mostly focus on the diagnostic performance of a single imaging technique. This study aims to construct a discrimination diagnostic model for NLM and BC based on such imaging features as ultrasound and magnetic resonance imaging (MRI) and to validate the application value of the model, assisting clinicians in improving disease diagnosis and refining medical decisions. Methods This study is a retrospective analysis. Clinical data of 108 patients suspected of NLM based on imaging diagnosis, admitted to The First Affiliated Hospital with Nanjing Medical University between May 2018 and August 2023, were collected. Among them, 94 cases were pathologically confirmed as NLM and 14 cases as BC. Univariate and multivariate logistic regression analyses were performed on the patients' clinical data, ultrasound features, and MRI features to select the risk factors for discriminating NLM and BC, and construct a discrimination model. The discrimination performance of the model was analyzed with the receiver operating characteristic (ROC) curve, decision curve analysis (DCA), and calibration curve. Results In the NLM group, there were 24 cases of granulomatous lobular mastitis (25.53%) and 70 cases of plasma cell mastitis (74.47%). In the BC group, there were 2 cases of infiltrating ductal carcinoma, 2 cases of atypical hyperplasia, 3 cases of papillary carcinoma, and 7 cases of ductal carcinoma in situ. Age, internal blood flow, calcification, edge, enhancement characteristics, apparent diffusion coefficient (ADC) values, and time-intensity curve (TIC) type were independent factors for differentiating NLM and BC (p < 0.05). The ROC analysis showed that the area under the curve of the model for discriminating NLM and BC was 0.920. The DCA results showed that the model had high net benefits for discriminating NLM and BC. The calibration curve analysis showed that the model had good consistency with the actual diagnosis of NLM and BC, with a chi-square value of 4.545 and a p-value of 0.155 according to the Hosmer-Lemeshow test. Conclusion Age, internal blood flow, calcification, edge, enhancement characteristics, ADC, and TIC curve types are important factors in distinguishing NLM and BC, and the model based on the above characteristics to distinguish NLM and BC has a high net benefit in distinguishing the two.

Artificial intelligence in cardiology: a machine learning model for supraventricular tachycardia discrimination

Abstract Background Distinguishing atrioventricular nodal reentrant tachycardia (AVNRT) from orthodromic atrioventricular reentrant tachycardia (AVRT) presents a diagnostic challenge, with clinical and electrocardiogram (ECG) features playing a crucial role in differential diagnosis. Purpose The aim of this study was to devise and validate a predictive algorithm that utilises basic clinical and ECG parameters to accurately classify AVNRT versus AVRT. Furthermore, this algorithm was integrated into a user-friendly, web-based tool to support clinical decision-making. Methods We conducted a prospective analysis of 704 patients (median age: 53.6 years; 59.7% female) with electrophysiologically confirmed AVNRT (n=535) or AVRT (n=169). A comprehensive evaluation was performed using a structured questionnaire tailored to arrhythmia-specific symptoms. The study utilised advanced machine learning techniques for classification, employing feature selection methods to enhance predictive accuracy. The analysis included 19 demographic, clinical, and ECG variables, adopting an 80-20% split for training and internal validation. The algorithm's performance was evaluated against a model comprising all variables (Full Model), expert ECG interpretations (ExpertECG), objective ECG criteria (ECG model), and a baseline dummy classifier, with efficacy assessed by the Area Under the Receiver Operating Characteristic curve (AUC) and the Area Under the Precision-Recall Curve (AUPRC). Results The optimised model, utilising the XGBoost algorithm with six critical variables (initial age at first episode, biological sex, presence of neck palpitations, pseudoR wave in lead V1/pseudoS wave in inferior leads, visibility of retrograde P-waves, and tachycardia cycle length), exhibited superior performance, achieving an AUC of 0.916 and an AUPRC of 0.823. This model significantly outperformed other comparative models in terms of predictive accuracy (Figure 1). Implementation of the model into a web-based interface was accomplished using Shiny, enhancing its accessibility for clinical application (Figure 2). Conclusions This study illustrates the efficacy of employing machine learning algorithms to utilise bedside clinical and ECG variables for accurately differentiating between AVNRT and AVRT. The creation of a web-based calculator marks a significant advancement, empowering clinicians, especially non-specialists, with a potent tool for providing personalised diagnostic insights. Future research should focus on external validation to ascertain the model's generalisability and its applicability in broader clinical settings.Figure 1Figure 2

Generative adversarial networks (GANs) to produce synthetic 12-lead electrocardiogram signals for specific and rare diseases: a novel, powerful tool towards clinical advancements

Abstract Background Remarkable technological advancements in electrocardiogram (ECG) analysis face challenges due to insufficiently diverse and extensive real-world datasets. This limitation hampers both the development of sophisticated deep learning (DL) models for ECG analysis and the exposure of new physicians to substantial training examples. Purpose To leverage Generative Adversarial Networks (GANs), a novel AI algorithm type, to generate synthetic yet realistic ECG signals resembling specific diseases. Methods We initially employed a general-purpose GAN on more than 21,000 ECG samples from the PTB-XL database, encompassing normal and abnormal ECGs across various diseases. The GAN outputs served as inputs for training "DeceptionECG", a novel GAN model inspired by the InceptionTime architecture (code available online). DeceptionECG focused on generating disease-specific ECG signals, trained on annotated ECGs validated by two electrophysiologists. Post-training, we generated synthetic ECGs for common diseases like atrial fibrillation (AF) and anterior ST-segment elevation myocardial infarction (STEMI), as well as rarer conditions like Wolf-Parkinson White (WPW) syndrome. Authentication involved an independent discriminator DL model and two cardiologists who were asked to distinguish real from synthetic ECGs. Additionally, a DL model, a variant of the InceptionTime architecture, discerned specific diseases in the ECG signals to assess their realism. Finally, we trained a DL model for detecting abnormalities in ECGs with the augmented dataset to evaluate potential performance improvement. Results We generated sets of 5,000 synthetic ECGs for various abnormalities, at a rate of 2,500 12-lead ECG signals per minute. The generated ECGs were indistinguishable from real ones, evidenced by low accuracy (Accuracy, ACC: 48.9%) of the discriminator model and two clinicians (ACC: 56% and 51%), on test sets with anterior STEMI. A DL model trained on real ECGs accurately differentiated between synthetic normal ECGs and those hosting the target disease in nearly 9 out of 10 cases (ACC: 88.7%). Furthermore, the employed DL model detector trained on the augmented dataset, encompassing both real (n=1,589) and synthetic (n=5,000) ECGs, outperformed its counterpart trained solely on real ECGs, when tested on a 1,000-sample set of real signals (ACC: 94.6% vs. 89.3%). Conclusion Our GAN-based approach showcases the potential of generating high-quality synthetic ECGs tailored for specific diseases, overcoming data scarcity challenges, and enhancing the training and performance of machine learning models. Additionally, it holds promise in advancing medical professional training by providing diverse signals with distinct "disease stamps" at massive amounts, addressing privacy concerns associated with extensive data exposure. Further testing is warranted for the applicability of the approach, especially in exceedingly rare conditions with limited available training data.Project workflow and outcomesGAN-generated ECGs: Which is fake?

Condition Monitoring Using a Latent Space of Variational Autoencoder Trained Only on a Healthy Machine.

Machine learning generative models have opened up a new perspective for automated machine diagnostics. These methods improve decision-making by extracting features, classifying, and creating new observations using deep neural networks. Generative modeling aims to determine the joint distribution of input data. This contrasts traditional methods used in diagnostics based on discriminative models and the conditional probability distribution of the target variable at known feature values. In the variational autoencoder (VAE) algorithms trained by the authors, the parameters of diagnostic features are random variables, the distributions of which can be approximated based on data, and the identification of probability distributions is based on variational inference. Variational inference is a tool that deals with difficult statistical problems and is usually faster than classical methods. VAEs can detect anomalies, predict failures, and optimize processes. This paper proposes an unsupervised approach to fault diagnosis using only healthy data with automatic feature extraction from the continuous probabilistic latent subspace of the VAE encoder and reduction in PCA or t-SNE. The solution, verified in the example of simulation data, is a response to a common problem related to the lack or difficulty of obtaining marked data in defected states of devices and mechanical structures.

Animal family discrimination from hair using ATR-FTIR and machine learning methods for applications in illegal wildlife trafficking.

Wildlife forensics plays a pivotal role in the combating illegal trafficking, supporting biodiversity conservation, and aiding in the identification of animals in wildlife. Animal hair, often found in trafficking crimes, serves as vital biological evidence that can provide significant information for animal identification. This study proposes a novel method integrating machine learning classifiers with Fourier transform infrared (FTIR) spectroscopy in attenuated total reflectance (ATR) mode to enhance the effectiveness of animal identification in wildlife forensic casework. Additionally, compound microscopy has also been utilized as a preliminary tool to perform morphological analysis of hair samples from four animal families, including Bovidae, Cervidae, Elephantidae, and Felidae. Further, chemical profiling through spectral data revealed significant overlapping peaks between family Bovidae and Cervidae. The classification experiment provides the random forest (RF) classifier as the most effective for family discrimination model. This research offers valuable insights for wildlife forensics by improving the identification accuracy of unknown hair samples, thus enhancing the overall effectiveness in forensic investigations.

Identification of drought-salinity combined stress in tomato plants by vegetation indices

A major issue in several farming areas of the Mediterranean basin consists of drought and salinity stress. This stress is mainly due to a steady exposition of warm daily temperature and heatwaves, moreover with inevitable irrigation with saline water. Therefore, detecting the stress is essential to minimise significant yield loss and preserve agricultural sustainability. In this context, remote and proximal sensing can play a crucial role in allowing fast, not destructive, extensive, and reliable assessment of crop status. In this work, the effectiveness of several multispectral indices in detecting salinity and water stress in tomato plants, grown under controlled green-house conditions, was investigated. Three different classifiers (fine tree model, linear discriminant model, and linear support vector machines model) were used to verify whether, and the extent to which, the adopted multispectral indices can be adopted to identify a stress condition of the tomato plants. In the experimental campaign, the stress occurrence on tomato plants was assessed on the base of a set of ecophysiological measurements, such as transpiration, stomatal conductance, and photosynthesis rate. Obtained results showed that a classification model based on linear support vector machines, exploiting the combination of Photochemical Reflectance Index and the Chlorophyl Index, can detect drought and salinity stress in tomato plants with an accuracy higher than 94%.