Independent Validation Dataset Research Articles

The Sleep Well Baby project: an automated real-time sleep-wake state prediction algorithm in preterm infants.

Study ObjectivesSleep is an important driver of early brain development. However, sleep is often disturbed in preterm infants admitted to the neonatal intensive care unit (NICU). We aimed to develop an automated algorithm based on routinely measured vital parameters to classify sleep–wake states of preterm infants in real-time at the bedside.MethodsIn this study, sleep–wake state observations were obtained in 1-minute epochs using a behavioral scale developed in-house while vital signs were recorded simultaneously. Three types of vital parameter data, namely, heart rate, respiratory rate, and oxygen saturation, were collected at a low-frequency sampling rate of 0.4 Hz. A supervised machine learning workflow was used to train a classifier to predict sleep–wake states. Independent training (n = 37) and validation datasets were validation n = 9) datasets were used. Finally, a setup was designed for real-time implementation at the bedside.ResultsThe macro-averaged area-under-the-receiver-operator-characteristic (AUROC) of the automated sleep staging algorithm ranged between 0.69 and 0.82 for the training data, and 0.61 and 0.78 for the validation data. The algorithm provided the most accurate prediction for wake states (AUROC = 0.80). These findings were well validated on an independent sample (AUROC = 0.77).ConclusionsWith this study, to the best of our knowledge, a reliable, nonobtrusive, and real-time sleep staging algorithm was developed for the first time for preterm infants. Deploying this algorithm in the NICU environment may assist and adapt bedside clinical work based on infants’ sleep–wake states, potentially promoting the early brain development and well-being of preterm infants.

Detecting Sleep and Nonwear in 24-h Wrist Accelerometer Data from the National Health and Nutrition Examination Survey.

Estimating physical activity, sedentary behavior, and sleep from wrist-worn accelerometer data requires reliable detection of sensor nonwear and sensor wear during both sleep and wake. This study aimed to develop an algorithm that simultaneously identifies sensor wake-wear, sleep-wear, and nonwear in 24-h wrist accelerometer data collected with or without filtering. Using sensor data labeled with polysomnography ( n = 21) and directly observed wake-wear data ( n = 31) from healthy adults, and nonwear data from sensors left at various locations in a home ( n = 20), we developed an algorithm to detect nonwear, sleep-wear, and wake-wear for "idle sleep mode" (ISM) filtered data collected in the 2011-2014 National Health and Nutrition Examination Survey. The algorithm was then extended to process original raw data collected from devices without ISM filtering. Both algorithms were further validated using a polysomnography-based sleep and wake-wear data set ( n = 22) and diary-based wake-wear and nonwear labels from healthy adults ( n = 23). Classification performance (F1 scores) was compared with four alternative approaches. The F1 score of the ISM-based algorithm on the training data set using leave-one-subject-out cross-validation was 0.95 ± 0.13. Validation on the two independent data sets yielded F1 scores of 0.84 ± 0.60 for the data set with sleep-wear and wake-wear and 0.94 ± 0.04 for the data set with wake-wear and nonwear. The F1 score when using original, raw data was 0.96 ± 0.08 for the training data sets and 0.86 ± 0.18 and 0.97 ± 0.04 for the two independent validation data sets. The algorithm performed comparably or better than the alternative approaches on the data sets. A novel machine-learning algorithm was designed to recognize wake-wear, sleep-wear, and nonwear in 24-h wrist-worn accelerometer data that are applicable for ISM-filtered data or original raw data.

A method for continuous rhythm classification and early detection of ventricular fibrillation during CPR

AimWe developed a method which continuously classifies the ECG rhythm during CPR in order to guide clinical care. MethodsWe conducted a retrospective study of 432 patients treated following out-of-hospital cardiac arrest. Continuous ECG sequences from two-minute CPR cycles were extracted from defibrillator recordings and further divided into five-second clips. We developed an algorithm using wavelet analysis, hidden semi-Markov modeling, and random forest classification. The algorithm classifies individual clips as asystole, organized rhythm, ventricular fibrillation, or Inconclusive while integrating information from previous clips. Classifications were compared to manual annotations to estimate accuracy in an independent validation dataset. Continuous sequences were classified as shockable, non-shockable, or Inconclusive; classifications were used to compute shock sensitivity and specificity. ResultsOf 432 patient-cases, 290 were used for development and 142 for validation. In the 12,294 validation ECG clips during CPR, accuracies were 0.88 (95% CI 0.85–0.91) for asystole, 0.98 (95% CI 0.98–0.99) for organized rhythm, and 0.97 (95% CI 0.96–0.97) for ventricular fibrillation, with 43% classified as Inconclusive. Of 457 continuous sequences, shock sensitivity was 0.90 (95% CI 0.86–0.93), shock specificity was 0.98 (95% CI 0.93–0.99), and 7% were Inconclusive. Median delay to ventricular fibrillation recognition was 10 (IQR 5–32) seconds. ConclusionA novel algorithm continuously classified the primary resuscitation rhythms—asystole, organized rhythms, and ventricular fibrillation—with 88–98% accuracy, enabling accurate shock advisory guidance during most two-minute CPR cycles. Additional investigation is required to understand how algorithm implementation could affect rescuer actions and clinical outcomes.

An Objective and Easy-to-Use Glaucoma Functional Severity Staging System Based on Artificial Intelligence.

The objective of this study was to develop an objective and easy-to-use glaucoma staging system based on visual fields (VFs). A total of 13,231 VFs from 8077 subjects were used to develop models and 8024 VFs from 4445 subjects were used to validate models. We developed an unsupervised machine learning model to identify clusters with similar VF values. We annotated the clusters based on their respective mean deviation (MD). We computed optimal MD thresholds that discriminate clusters with the highest accuracy based on Bayes minimum error principle. We evaluated the accuracy of the staging system and validated findings based on an independent validation dataset. The unsupervised k -means algorithm discovered 4 clusters with 6784, 4034, 1541, and 872 VFs and average MDs of 0.0dB (±1.4: SD), -4.8dB (±1.9), -12.2dB (±2.9), and -23.0dB (±3.8), respectively. The supervised Bayes minimum error classifier identified optimal MD thresholds of -2.2, -8.0, and -17.3dB for discriminating normal eyes and eyes at the early, moderate, and advanced stages of glaucoma. The accuracy of the glaucoma staging system was 94%, based on identified MD thresholds with respect to the initial k -means clusters. We discovered that 4 severity levels based on MD thresholds of -2.2, -8.0, and -17.3dB, provides the optimal number of severity stages based on unsupervised and supervised machine learning. This glaucoma staging system is unbiased, objective, easy-to-use, and consistent, which makes it highly suitable for use in glaucoma research and for day-to-day clinical practice.

Construction of a hypoxic immune microenvironment associated gene-based model for prognosis prediction of lung adenocarcinoma.

Lung adenocarcinoma (LUAD) accounts for the majority of cancer deaths worldwide, with a high incidence rate and mortality. It is highly important to develop biomarker model to accurately predict the prognosis. RNA-Seq data and clinical follow-up data of LUAD were downloaded from The Cancer Genome Atlas (TCGA) database. Hypoxia-related gene sets were collected from the Gene Set Enrichment Analysis (GSEA) website. A gene signature model was established using the Limma package in the R software, univariate and multivariate survival analyses, and least absolute shrinkage and selection operator (LASSO) algorithms. Two hypoxia subtypes (C1 and C2) were classified according to the expressions of 55 prognostic hypoxic-related genes. Differentially expressed genes (DEGs) between two hypoxia subtypes and immune group were analyzed. Then, 390 DEGs related to hypoxic immune microenvironment were filtered. According to hypoxia type and immune type, the samples were classified into hypoxia-high & immune-low group, hypoxia-low & immune-high group. Based on these differentially expressed genes (DEGs), a 5-genes signature model, which showed a stable prediction performance on datasets of different platforms and immunotherapy datasets, was finally developed. Meanwhile, it demonstrated a better performance compared with other existing models. The AUC of the 5-gene signature was high in both the training dataset and 4 independent validation datasets and was confirmed as a clinical feature-independent prognostic model. This study developed a hypoxic immune microenvironment associated gene-based model for prognostic prediction of LUAD, providing clinicians with a reliable prognostic assessment tool and facilitating clinical treatment decision-making.

Are polymorphisms affecting serum urate, renal urate handling and alcohol intake associated with co-morbidities in gout cases? A case–control study using data from the UK Biobank

To examine the association between common comorbidities, eGFR and loci involved in the hyperuricaemia-gout transition. This study was conducted in people with gout from the UK Biobank. Logistic regression was used to examine the association between self-reported physician-diagnosed hypertension, diabetes, hypercholesterolemia and ischaemic heart disease (IHD) with the following variants: rs1260326(GCKR), rs16890979(SLC2A9), rs2231142(ABCG2), rs1229984(ADH1B) and rs2078267(SLC22A11) and adjusted for age, sex and 10-principal components. Linear regression was used to examine the association with eGFR. 7,049 participants with gout were included. After adjusting for multiple testing, there was a statistically significant positive association between urate lowering allele at SLC2A9 and hypertension, and negative association between urate raising allele at ABCG2 and hypertension (OR 1.17 and OR 0.86, respectively). Number of urate lowering risk alleles associated with hypertension [OR (95%CI) 1.13 (1.06–1.21)]. High eGFR associated with urate raising allele at rs2231142 (β = 1.38). The SNP in ADH1B that protects from alcohol excess showed a negative association with IHD (OR 0.53). Unlike in general population studies urate lowering genetic variants associate with hypertension in gout patients with dose–response. This may be due to high prevalence of other risk factors of hypertension such as obesity, poor diet etc. and needs validation in independent datasets.

Exploring and Identifying Prognostic Phenotypes of Patients with Heart Failure Guided by Explainable Machine Learning.

Identifying patient prognostic phenotypes facilitates precision medicine. This study aimed to explore phenotypes of patients with heart failure (HF) corresponding to prognostic condition (risk of mortality) and identify the phenotype of new patients by machine learning (ML). A unsupervised ML was applied to explore phenotypes of patients in a derivation dataset (n = 562) based on their medical records. Thereafter, supervised ML models were trained on the derivation dataset to classify these identified phenotypes. Then, the trained classifiers were further validated on an independent validation dataset (n = 168). Finally, Shapley additive explanations were used to interpret decision making of phenotype classification. Three patient phenotypes corresponding to stratified mortality risk (high, low, and intermediate) were identified. Kaplan–Meier survival curves among the three phenotypes had significant difference (pairwise comparison p < 0.05). Hazard ratio of all-cause mortality between patients in phenotype 1 (n = 91; high risk) and phenotype 3 (n = 329; intermediate risk) was 2.08 (95%CI 1.29–3.37, p = 0.003), and 0.26 (95%CI 0.11–0.61, p = 0.002) between phenotype 2 (n = 142; low risk) and phenotype 3. For phenotypes classification by random forest, AUCs of phenotypes 1, 2, and 3 were 0.736 ± 0.038, 0.815 ± 0.035, and 0.721 ± 0.03, respectively, slightly better than the decision tree. Then, the classifier effectively identified the phenotypes for new patients in the validation dataset with significant difference on survival curves and hazard ratios. Finally, age and creatinine clearance rate were identified as the top two most important predictors. ML could effectively identify patient prognostic phenotypes, facilitating reasonable management and treatment considering prognostic condition.

PeakBot: machine-learning-based chromatographic peak picking.

MotivationChromatographic peak picking is among the first steps in data processing workflows of raw LC-HRMS datasets in untargeted metabolomics applications. Its performance is crucial for the holistic detection of all metabolic features as well as their relative quantification for statistical analysis and metabolite identification. Random noise, non-baseline separated compounds and unspecific background signals complicate this task.ResultsA machine-learning-based approach entitled PeakBot was developed for detecting chromatographic peaks in LC-HRMS profile-mode data. It first detects all local signal maxima in a chromatogram, which are then extracted as super-sampled standardized areas (retention-time versus m/z). These are subsequently inspected by a custom-trained convolutional neural network that forms the basis of PeakBot’s architecture. The model reports if the respective local maximum is the apex of a chromatographic peak or not as well as its peak center and bounding box. In training and independent validation datasets used for development, PeakBot achieved a high performance with respect to discriminating between chromatographic peaks and background signals (accuracy of 0.99). For training the machine-learning model a minimum of 100 reference features are needed to learn their characteristics to achieve high-quality peak-picking results for detecting such chromatographic peaks in an untargeted fashion. PeakBot is implemented in python (3.8) and uses the TensorFlow (2.5.0) package for machine-learning related tasks. It has been tested on Linux and Windows OSs.Availability and implementationThe package is available free of charge for non-commercial use (CC BY-NC-SA). It is available at https://github.com/christophuv/PeakBot.Supplementary information Supplementary data are available at Bioinformatics online.

A Novel Cuproptosis-Related Prognostic Gene Signature and Validation of Differential Expression in Clear Cell Renal Cell Carcinoma.

Clear cell renal cell carcinoma (ccRCC) is the most prevalent subtype of renal cell carcinoma, which is characterized by metabolic reprogramming. Cuproptosis, a novel form of cell death, is highly linked to mitochondrial metabolism and mediated by protein lipoylation. However, the clinical impacts of cuproptosis-related genes (CRGs) in ccRCC largely remain unclear. In the current study, we systematically evaluated the genetic alterations of cuproptosis-related genes in ccRCC. Our results revealed that CDKN2A, DLAT, DLD, FDX1, GLS, PDHA1 and PDHB exhibited differential expression between ccRCC and normal tissues (|log2(fold change)| > 2/3 and p < 0.05). Utilizing an iterative sure independence screening (SIS) method, we separately constructed the prognostic signature of CRGs for predicting the overall survival (OS) and progression-free survival (PFS) in ccRCC patients. The prognostic score of CRGs yielded an area under the curve (AUC) of 0.658 and 0.682 for the prediction of 5-year OS and PFS, respectively. In the Kaplan−Meier survival analysis of OS, a higher risk score of cuproptosis-related gene signature was significantly correlated with worse overall survival (HR = 2.72 (2.01–3.68), log-rank p = 1.76 × 10−7). Patients with a higher risk had a significantly shorter PFS (HR = 2.83 (2.08–3.85), log-rank p = 3.66 × 10−7). Two independent validation datasets (GSE40435 (N = 101), GSE53757 (N = 72)) were collected for meta-analysis, suggesting that CDKN2A (log2(fold change) = 1.46, 95%CI: 1.75–2.35) showed significantly higher expression in ccRCC tissues while DLAT (log2(fold change) = −0.54, 95%CI: −0.93–−0.15) and FDX1 (log2(fold change) = −1.01, 95%CI: −1.61–−0.42) were lowly expressed. The expression of CDKN2A and FDX1 in ccRCC was also significantly associated with immune infiltration levels and programmed cell death protein 1 (PD-1) expression (CDKN2A: r = 0.24, p = 2.14 × 10−8; FDX1: r = −0.17, p = 1.37 × 10−4). In conclusion, the cuproptosis-related gene signature could serve as a potential prognostic predictor for ccRCC patients and may offer novel insights into the cancer treatment.

Unpaired Cross-Modality Educed Distillation (CMEDL) for Medical Image Segmentation.

Accurate and robust segmentation of lung cancers from CT, even those located close to mediastinum, is needed to more accurately plan and deliver radiotherapy and to measure treatment response. Therefore, we developed a new cross-modality educed distillation (CMEDL) approach, using unpaired CT and MRI scans, whereby an informative teacher MRI network guides a student CT network to extract features that signal the difference between foreground and background. Our contribution eliminates two requirements of distillation methods: (i) paired image sets by using an image to image (I2I) translation and (ii) pre-training of the teacher network with a large training set by using concurrent training of all networks. Our framework uses an end-to-end trained unpaired I2I translation, teacher, and student segmentation networks. Architectural flexibility of our framework is demonstrated using 3 segmentation and 2 I2I networks. Networks were trained with 377 CT and 82 T2w MRI from different sets of patients, with independent validation (N = 209 tumors) and testing (N = 609 tumors) datasets. Network design, methods to combine MRI with CT information, distillation learning under informative (MRI to CT), weak (CT to MRI) and equal teacher (MRI to MRI), and ablation tests were performed. Accuracy was measured using Dice similarity (DSC), surface Dice (sDSC), and Hausdorff distance at the 95th percentile (HD95). The CMEDL approach was significantly (p < 0.001) more accurate (DSC of 0.77 vs. 0.73) than non-CMEDL methods with an informative teacher for CT lung tumor, with a weak teacher (DSC of 0.84 vs. 0.81) for MRI lung tumor, and with equal teacher (DSC of 0.90 vs. 0.88) for MRI multi-organ segmentation. CMEDL also reduced inter-rater lung tumor segmentation variabilities.

DeepHPI: a comprehensive deep learning platform for accurate prediction and visualization of host-pathogen protein-protein interactions.

Host-pathogen protein interactions (HPPIs) play vital roles in many biological processes and are directly involved in infectious diseases. With the outbreak of more frequent pandemics in the last couple of decades, such as the recent outburst of Covid-19 causing millions of deaths, it has become more critical to develop advanced methods to accurately predict pathogen interactions with their respective hosts. During the last decade, experimental methods to identify HPIs have been used to decipher host-pathogen systems with the caveat that those techniques are labor-intensive, expensive and time-consuming. Alternatively, accurate prediction of HPIs can be performed by the use of data-driven machine learning. To provide a more robust and accurate solution for the HPI prediction problem, we have developed a deepHPI tool based on deep learning. The web server delivers four host-pathogen model types: plant-pathogen, human-bacteria, human-virus and animal-pathogen, leveraging its operability to a wide range of analyses and cases of use. The deepHPI web tool is the first to use convolutional neural network models for HPI prediction. These models have been selected based on a comprehensive evaluation of protein features and neural network architectures. The best prediction models have been tested on independent validation datasets, which achieved an overall Matthews correlation coefficient value of 0.87 for animal-pathogen using the combined pseudo-amino acid composition and conjoint triad (PAAC_CT) features, 0.75 for human-bacteria using the combined pseudo-amino acid composition, conjoint triad and normalized Moreau-Broto feature (PAAC_CT_NMBroto), 0.96 for human-virus using PAAC_CT_NMBroto and 0.94 values for plant-pathogen interactions using the combined pseudo-amino acid composition, composition and transition feature (PAAC_CTDC_CTDT). Our server running deepHPI is deployed on a high-performance computing cluster that enables large and multiple user requests, and it provides more information about interactions discovered. It presents an enriched visualization of the resulting host-pathogen networks that is augmented with external links to various protein annotation resources. We believe that the deepHPI web server will be very useful to researchers, particularly those working on infectious diseases. Additionally, many novel and known host-pathogen systems can be further investigated to significantly advance our understanding of complex disease-causing agents. The developed models are established on a web server, which is freely accessible at http://bioinfo.usu.edu/deepHPI/.

Evaluation of a One-Dimensional Convolution Neural Network for Chlorophyll Content Estimation Using a Compact Spectrometer

Leaf chlorophyll content is used as a major indicator of plant stress and growth, and hyperspectral remote sensing is frequently used to monitor the chlorophyll content. Hyperspectral reflectance has been used to evaluate vegetation properties such as pigment content, plant structure and physiological features using portable spectroradiometers. However, the prices of these devices have not yet decreased to consumer-affordable levels, which prevents widespread use. In this study, a system based on a cost-effective fingertip-sized spectrometer (Colorcompass-LF, a total price for the proposed solution was approximately 1600 USD) was evaluated for its ability to estimate the chlorophyll contents of radish and wasabi leaves and was compared with the Analytical Spectral Devices FieldSpec4. The chlorophyll contents per leaf area (cm2) of radish were generally higher than those of wasabi and ranged from 42.20 to 94.39 μg/cm2 and 11.39 to 40.40 μg/cm2 for radish and wasabi, respectively. The chlorophyll content was estimated using regression models based on a one-dimensional convolutional neural network (1D-CNN) that was generated after the original reflectance from the spectrometer measurements was de-noised. The results from an independent validation dataset confirmed the good performance of the Colorcompass-LF after spectral correction using a second-degree polynomial, and very similar estimation accuracies were obtained for the measurements from the FieldSpec4. The coefficients of determination of the regression models based on 1D-CNN were almost same (with R2 = 0.94) and the ratios of performance to deviation based on reflectance after spectral correction using a second-degree polynomial for the Colorcompass-LF and the FieldSpec4 were 4.31 and 4.33, respectively.

Mid-Infrared Reflectance Spectroscopy for Estimation of Soil Properties of Alfisols from Eastern India

Mid-infrared (MIR) spectroscopy is emerging as one of the most promising technologies, as it is a rapid and cost-effective alternative to routine laboratory analysis for many soil properties. This study was conducted to evaluate the potential of mid-infrared spectroscopy for the rapid and nondestructive measurement of some important soil properties of Alfisols. A total of 336 georeferenced soil samples fromthe 0–15 cm soil layer of Alfisols that were collected from the eastern Indian states of Odisha and Jharkhand were used. The partial least-squares regression (PLSR), random forest, and support vector machine regression techniques were compared for the calibration of the spectral data with the wet chemistry soil data. The PLSR-based predictive models performed better than the other two regression techniques for all the soil properties, except for the electrical conductivity (EC). Good predictions with independent validation datasets were obtained for the clay and sand percentages and for the soil organic carbon (SOC) content, while satisfactory predictions were achieved for the silt percentage and the pH value. However, the performance of the predictive models was poor in the case of the EC and the extractable nutrients, such as the available phosphorus and potassium contents of the soil. Specific regions of the MIR spectra that contributed to the prediction of the soil SOC, the pH, and the clay and sand percentages were identified. The study demonstrates the potential of the MIR spectroscopic technique in the simultaneous estimation of the SOC content, the sand, clay, and silt percentages, and the pH of Alfisols from eastern India.

Outcome-Based Decision-Making Algorithm for Treating Patients with Primary Aldosteronism.

BackgroundOptimal management of primary aldosteronism (PA) is crucial due to the increased risk of cardiovascular and cerebrovascular diseases. Adrenal venous sampling (AVS) is the gold standard method for determining subtype but is technically challenging and invasive. Some PA patients do not benefit clinically from surgery. We sought to develop an algorithm to improve decision-making before engaging in AVS and surgery in clinical practice.MethodsWe conducted the ongoing Korean Primary Aldosteronism Study at two tertiary centers. Study A involved PA patients with successful catheterization and a unilateral nodule on computed tomography and aimed to predict unilateral aldosterone-producing adenoma (n=367). Study B involved similar patients who underwent adrenalectomy and aimed to predict postoperative outcome (n=330). In study A, we implemented important feature selection using the least absolute shrinkage and selection operator regression.ResultsWe developed a unilateral PA prediction model using logistic regression analysis: lowest serum potassium level ≤3.4 mEq/L, aldosterone-to-renin ratio ≥150, plasma aldosterone concentration ≥30 ng/mL, and body mass index <25 kg/m2 (area under the curve, 0.819; 95% confidence interval, 0.774 to 0.865; sensitivity, 97.6%; specificity, 25.5%). In study B, we identified female, hypertension duration <5 years, anti-hypertension medication <2.5 daily defined dose, and the absence of coronary artery disease as predictors of clinical success, using stepwise logistic regression models (sensitivity, 94.2%; specificity, 49.3%). We validated our algorithm in the independent validation dataset (n=53).ConclusionWe propose this new outcome-driven diagnostic algorithm, simultaneously considering unilateral aldosterone excess and clinical surgical benefits in PA patients.

Control of a PVT-Heat-Pump-System Based on Reinforcement Learning–Operating Cost Reduction through Flow Rate Variation

This study aims to develop a controller to operate an energy system-consisting of a photovoltaic thermal (PVT) system combined with a heat pump, using the reinforcement learning approach to minimize the operating costs of the system. For this, the flow rate of the cooling fluid pumped through the PVT system is controlled. This flow rate determines the temperature increase of the cooling fluid while reducing the temperature of the PVT system. The heated-up cooling fluid is used to improve the heat pump’s coefficient of performance (COP). For optimizing the operation costs of such a system, first an extensive simulation model has been developed. Based on this technical model, a controller has been developed using the reinforcement learning approach to allow for a cost-efficient control of the flow rate. The results show that a successfully trained control unit based on the reinforcement learning approach can reduce the operating costs with an independent validation dataset. For the case study presented here, based on the implemented methodological approach, including hyperparameter optimization, the operating costs of the investigated energy system can be reduced by more than 4% in the training dataset and by close to 3% in the validation dataset.

Laser desorption tissue imaging with Differential Mobility Spectrometry

Pathological gross examination of breast carcinoma samples is sometimes laborious. A tissue pre-mapping method could indicate neoplastic areas to the pathologist and enable focused sampling. Differential Mobility Spectrometry (DMS) is a rapid and affordable technology for complex gas mixture analysis. We present an automated tissue laser analysis system for imaging approaches (iATLAS), which utilizes a computer-controlled laser evaporator unit coupled with a DMS gas analyzer. The system is demonstrated in the classification of porcine tissue samples and three human breast carcinomas. Tissue samples from eighteen landrace pigs were classified with the system based on a pre-designed matrix (spatial resolution 1–3 mm). The smoke samples were analyzed with DMS, and tissue classification was performed with several machine learning approaches. Porcine skeletal muscle (n = 1030), adipose tissue (n = 1329), normal breast tissue (n = 258), bone (n = 680), and liver (n = 264) were identified with 86% cross-validation (CV) accuracy with a convolutional neural network (CNN) model. Further, a panel tissue that comprised all five tissue types was applied as an independent validation dataset. In this test, 82% classification accuracy with CNN was achieved. An analogous procedure was applied to demonstrate the feasibility of iATLAS in breast cancer imaging according to 1) macroscopically and 2) microscopically annotated data with 10-fold CV and SVM (radial kernel). We reached a classification accuracy of 94%, specificity of 94%, and sensitivity of 93% with the macroscopically annotated data from three breast cancer specimens. The microscopic annotation was applicable to two specimens. For the first specimen, the classification accuracy was 84% (specificity 88% and sensitivity 77%). For the second, the classification accuracy was 72% (specificity 88% and sensitivity 24%). This study presents a promising method for automated tissue imaging in an animal model and lays foundation for breast cancer imaging.

Analysis of potential genetic biomarkers using machine learning methods and immune infiltration regulatory mechanisms underlying atrial fibrillation

ObjectiveWe aimed to screen out biomarkers for atrial fibrillation (AF) based on machine learning methods and evaluate the degree of immune infiltration in AF patients in detail.MethodsTwo datasets (GSE41177 and GSE79768) related to AF were downloaded from Gene expression omnibus (GEO) database and merged for further analysis. Differentially expressed genes (DEGs) were screened out using “limma” package in R software. Candidate biomarkers for AF were identified using machine learning methods of the LASSO regression algorithm and SVM-RFE algorithm. Receiver operating characteristic (ROC) curve was employed to assess the diagnostic effectiveness of biomarkers, which was further validated in another independent validation dataset of GSE14975. Moreover, we used CIBERSORT to study the proportion of infiltrating immune cells in each sample, and the Spearman method was used to explore the correlation between biomarkers and immune cells.Results129 DEGs were identified, and CYBB, CXCR2, and S100A4 were identified as key biomarkers of AF using LASSO regression and SVM-RFE algorithm. Both in the training dataset and the validation dataset, CYBB, CXCR2, and S100A4 showed favorable diagnostic effectiveness. Immune infiltration analysis indicated that, compared with sinus rhythm (SR), the atrial samples of patients with AF contained a higher T cells gamma delta, neutrophils and mast cells resting, whereas T cells follicular helper were relatively lower. Correlation analysis demonstrated that CYBB, CXCR2, and S100A4 were significantly correlated with the infiltrating immune cells.ConclusionsIn conclusion, this study suggested that CYBB, CXCR2, and S100A4 are key biomarkers of AF correlated with infiltrating immune cells, and infiltrating immune cells play pivotal roles in AF.

Transcriptomic Biomarker Signatures for Discrimination of Oral Cancer Surgical Margins.

Relapse after surgery for oral squamous cell carcinoma (OSCC) contributes significantly to morbidity, mortality and poor outcomes. The current histopathological diagnostic techniques are insufficiently sensitive for the detection of oral cancer and minimal residual disease in surgical margins. We used whole-transcriptome gene expression and small noncoding RNA profiles from tumour, close margin and distant margin biopsies from 18 patients undergoing surgical resection for OSCC. By applying multivariate regression algorithms (sPLS-DA) suitable for higher dimension data, we objectively identified biomarker signatures for tumour and marginal tissue zones. We were able to define molecular signatures that discriminated tumours from the marginal zones and between the close and distant margins. These signatures included genes not previously associated with OSCC, such as MAMDC2, SYNPO2 and ARMH4. For discrimination of the normal and tumour sampling zones, we were able to derive an effective gene-based classifying model for molecular abnormality based on a panel of eight genes (MMP1, MMP12, MYO1B, TNFRSF12A, WDR66, LAMC2, SLC16A1 and PLAU). We demonstrated the classification performance of these gene signatures in an independent validation dataset of OSCC tumour and marginal gene expression profiles. These biomarker signatures may contribute to the earlier detection of tumour cells and complement existing surgical and histopathological techniques used to determine clear surgical margins.

Automated diagnosis of age-related macular degeneration using multi-modal vertical plane feature fusion via deep learning.

To develop a computer-aided diagnostic (CADx) system of age-related macular degeneration (AMD) through feature fusion between infrared reflectance (IR) and optical coherence tomography (OCT) modalities in order to explore the superiority of multi-modality CADx system and the optimal feature fusion patterns between multi-modality inputs. This is a dual center retrospective study. We retrospectively collected 2006 pairs of IR and OCT images to develop the algorithms. Two single-modality models and three multi-modality models were constructed for the comparison of the diagnostic efficacy. The multi-modality models were designed utilizing a novel feature fusion method, namely, vertical plane feature fusion (VPFF). The results were validated using an independent external validation dataset and compared by three ophthalmologists. In the test set of the ZJU dataset, our best model named OCT_MAIN demonstrated diagnostic efficiency with an overall accuracy of 0.9608 and area under the curve of 0.9944 for the normal category, 0.9659 for the dry AMD category, and 0.9930 for the wet AMD category. The external validation exhibited an overall accuracy of 0.9159. Its diagnostic efficiency was comparable to that of the senior ophthalmologist. The VPFF method was successfully employed to develop a multi-modal intelligent diagnostic system for the AMD classification. This is a valuable complement and optimization to the existing CADx system, which reveals a wide application prospect and research potential.

Evaluating Sentinel-2 red edge through hyperspectral profiles for monitoring LAI & chlorophyll content of Kinnow Mandarin orchards

Leaf area index (LAI) and chlorophyll content are efficient plant health and nutrition indicators. However, retrieving these traits is challenging for the multilayered canopy of evergreen fruit trees like Citrus. The spectral reflectance's Red Edge Position (REP) is a surrogate measure of LAI and chlorophyll content. Sentinel-2 (20 m) estimates REP at a much higher spatial resolution than before. However, retrieval of the traits in the dense canopy structure of orchards requires proper validation of the Sentinel-2 REP (S2REP) through the hyperspectral profiles of plants. The objective here is to evaluate the potential of S2REP to retrieve the LAI and chlorophyll of kinnow mandarin fruit orchards and validate the results with REP extracted from the hyperspectral profiles and ground measurements of these traits. For this, we first tested several algorithms to enumerate the REP range from hyperspectral data that corresponds well to the kinnow mandarin's LAI and Chl content. Next, we used the S2REP to estimate these traits at sampled locations. Hence, we calibrated the regression models of LAI and chlorophyll content with the Sentinel-2 REP as an explanatory variable and traits ground measurements as response variables. The prediction accuracy was tested through an independent validation data set. The results show that the kinnow mandarin trees demonstrated significant spectral response in the 695–725 nm S2REP range of the electromagnetic spectrum similar to the spectral profiles from ground-based hyperspectral data. With this S2REP range, LAI and chlorophyll content models outperformed to predict these traits for the orchards, with adjusted-R2 0.86 and 0.80 and RMSE 3.7 and 53.3%, respectively. We conclude that the S2REP can effectively evaluate kinnow mandarin's LAI and Chlorophyll over large areas. The plants' LAI and chlorophyll content are best represented by the REP derived from the hyperspectral profiles through Polynomial fitting and Linear Extrapolation. Such spectrally-validated Sentinel-2 REP for kinnow orchards can calibrate the LAI and chlorophyll models more accurately over large scales. Our methods can be extrapolated to other fruit traits to enable the real-time monitoring of orchards' health and nutritional status.