Variations In Features Research Articles

Comparison of leaf chlorophyll content retrieval performance of citrus using FOD and CWT methods with field-based full-spectrum hyperspectral reflectance data

Citrus is one of the most economically valuable fruit trees in the world, for which leaf chlorophyll content (LCC) serves as a crucial indicator for evaluating its growth and health status. However, the quantitative estimation of LCC using remote-sensing techniques is still challenging owing to unclear sensitive spectral ranges, baseline drift and overlapping spectrum peaks. To resolve these issues, we clarified the spectral response characteristics of citrus LCC using fractional-order derivatives (FOD) and continuous wavelet transform (CWT) methods to determine its sensitive spectral range with in situ full-spectrum leaf hyperspectral data. We proposed a novel method for estimating the LCC of citrus by combining an ensemble learning regression model based on Hyperopt optimization (H-ELR) with partial least squares regression (PLSR) using the ultra-high dimensional feature variables produced by dual- and tri-band combination strategies. We evaluated the retrieval performance of LCC between the FOD- and CWT-based optimal spectral feature variables. Besides, we further examined the feasibility of improving the estimation accuracy of LCC by the combination of their optimal feature variables. Finally, we evaluated the effect of the spectral curve trend changes and different dimensional spectral features on LCC estimation. The results showed that: (1) The FOD- and CWT-based methods improved the correlation between original spectral reflectance and LCC, with the correlation coefficient increasing by 0.046 and 0.054, respectively. We confirmed that 425–740 nm is the optimal spectral range for LCC estimation. (2) We found that the 0.9 order derivative Tri-band index (TBI2 (R571, R1697, R740)) constructed by combining the sensitive spectral bands of leaf water content achieved a high-precision LCC estimation (R2 = 0.876). In addition, the MERIS terrestrial chlorophyll index (MTCI) constructed based on the scale6 of CWT also gained good retrieval accuracy (R2 = 0.806). (3) We demonstrated that a combination of FOD-based and CWT-based sensitive spectral features improves estimation accuracies (R2 = 0.891) of LCC and revealed that the reflectance peaks and slope peaks at 550 and 750 nm are essential variables for predicting the citrus LCC. (4) The combination of PLSR and H-ELR model provided a good retrieval performance of citrus LCC with the kurtosis (γ = 3.2) and skewness (Sk = 0.066) of the residual prediction values. Our proposed method can provide a scientific basis for estimating LCC and other physiological parameters of citrus and other crop types, which is also important to optimize agricultural management practices and improve crop yield.

Mapping green innovation with machine learning: Evidence from China

Achieving green innovation to eliminate the environmental impact of economic activities has been an important task for firms across industries. This study uses the resource-based view (RBV) and upper echelons theory (UET) to develop a green innovation determinants model and explore how firm characteristics, chief executive officer (CEO) characteristics, environmental actions, environmental disclosures, and industry characteristics influence corporate green innovation. Using a large-scale dataset of Chinese companies from 2010 to 2019 to identify the determinants of green innovation through machine learning algorithms, our results suggest that the extreme gradient boosting (XGBoost) method is the best model for predicting green innovation. After that, we visualized the effects and importance of the various feature variables through the best prediction model. Our results indicate that the environmental, social, and governance (ESG) rating is the most crucial determinant for green innovation, followed by internationalization, CEO pay, firm sales, industry size, research and development (R&D) intensity, and CEO education. Overall, our findings contribute to a broader understanding of the drivers of green innovation and offer critical implications for managers and policymakers to improve sustainable development.

Multi-scale multi-attention network for diabetic retinopathy grading

Objective. Diabetic retinopathy (DR) grading plays an important role in clinical diagnosis. However, automatic grading of DR is challenging due to the presence of intra-class variation and small lesions. On the one hand, deep features learned by convolutional neural networks often lose valid information about these small lesions. On the other hand, the great variability of lesion features, including differences in type and quantity, can exhibit considerable divergence even among fundus images of the same grade. To address these issues, we propose a novel multi-scale multi-attention network (MMNet). Approach. Firstly, to focus on different lesion features of fundus images, we propose a lesion attention module, which aims to encode multiple different lesion attention feature maps by combining channel attention and spatial attention, thus extracting global feature information and preserving diverse lesion features. Secondly, we propose a multi-scale feature fusion module to learn more feature information for small lesion regions, which combines complementary relationships between different convolutional layers to capture more detailed feature information. Furthermore, we introduce a Cross-layer Consistency Constraint Loss to overcome semantic differences between multi-scale features. Main results. The proposed MMNet obtains a high accuracy of 86.4% and a high kappa score of 88.4% for multi-class DR grading tasks on the EyePACS dataset, while 98.6% AUC, 95.3% accuracy, 92.7% recall, 95.0% precision, and 93.3% F1-score for referral and non-referral classification on the Messidor-1 dataset. Extensive experiments on two challenging benchmarks demonstrate that our MMNet achieves significant improvements and outperforms other state-of-the-art DR grading methods. Significance. MMNet has improved the diagnostic efficiency and accuracy of diabetes retinopathy and promoted the application of computer-aided medical diagnosis in DR screening.

Fine-Scale Spatial Variability of Stand Structural Features under Selection Management and Strict Protection: An Example from the Dinaric Mountains

Small scattered plots of a few hundred square meters well reflect structural variability at stand level, but not at small spatial scales as the data between plots is missing. Information about structural similarities between managed and unmanaged stands, especially based on large sample plots, is still scarce. Our first objective was to quantify and illustrate structural variability of a selection-managed stand and a corresponding old-growth (OG) stand at small spatial scales. The second goal was to find out if there is a positive autocorrelation among neighboring patches in these stands regarding tree density (N) and basal area (BA). Tree positions and their diameters were recorded in 1.5 ha plots. Structural variation was examined at scales from 0.01 ha to 0.36 ha. Spatial correlation of N and BA was examined by applying experimental semivariograms. The variability of N was similar in both stands, whereas it significantly differed regarding BA (α = 0.05). Semivariance did not detect positive spatial autocorrelation of BA, while adjacent plots appeared to be more similar (autocorrelated) regarding N in both stands. Despite statistical difference regarding BA variability, the selection-managed stand exhibited many structural similarities to the OG stand, which makes it potentially suitable for modulating, if needed, to bring it step closer to an old-growth structure.

Corroded submarine pipeline degradation prediction based on theory-guided IMOSOA-EL model

Submarine pipelines play an important role in the oil and gas transportation. However, pipeline corrosion can cause the structural degradation of pipelines, which may lead to failure accident and environmental pollution. Pipeline degradation prediction based on monitoring data is of great significance for preventing corrosion failure. In this paper, an ensemble learning (EL) based approach integrated with multi-objective optimization is developed for pipeline corrosion degradation prediction. Firstly, the engineering theory and domain knowledge are integrated into feature engineering to improve the interpretability. Several new feature variables are constructed based on the corrosion mechanism and empirical models. Secondly, an improved multi-objective seagull optimization algorithm (IMOSOA) is proposed to optimize the hyper-parameters of EL model. Subsequently, a novel data-driven model, so-called theory-guided IMOSOA-EL is proposed for the corrosion rate prediction. And six benchmark functions are used to verify the optimization performance of IMOSOA. Then, different feature subsets are developed based on correlation analysis. A comprehensive evaluation indicator is proposed for the optimal feature subset selection. The results demonstrate that the proposed model presents superiority in prediction accuracy compared with other models. This study is significant for reliability assessment and maintenance decision-making of submarine pipelines.

Fault Diagnosis of Oil-Immersed Transformers Based on the Improved Neighborhood Rough Set and Deep Belief Network

As one of the essential components in power systems, transformers play a pivotal role in the transmission and distribution of renewable energy generation. Accurate diagnosis of transformer fault types is crucial for maintaining the safety of power systems. The current focus in research lies in transformer fault diagnosis methods based on Dissolved Gas Analysis (DGA). Traditional diagnostic methods directly utilize the five fault gases from DGA data as model input features, but this approach does not comprehensively reflect all potential fault types in transformers. In this paper, a non-coding ratio method was employed to generate 35 fault gas ratios based on the five fault gases, subsequently refined through correlation analysis to eliminate redundant feature variables, resulting in 15 significantly representative fault gas ratios. To further streamline the feature variables and remove non-contributing elements to fault diagnosis, an improved Neighborhood Rough Set (INRS) algorithm was introduced, leveraging symmetrical uncertainty measurement. By resorting to the proposed INRS, eight most representative fault gas ratios were selected as input variables for constructing a Deep Belief Network (DBN) diagnostic model. Experimental results on Dissolved Gas Analysis (DGA) data confirmed the effectiveness and accuracy of the proposed method.

Evaluation of Flavor Type of Tobacco Blending Module: A Prediction Model Based on Near-Infrared Spectrum

Near-infrared spectrum technology is extensively employed in assessing the quality of tobacco blending modules, which serve as the fundamental units of cigarette production. This technology provides valuable technical support for the scientific evaluation of these modules. In this study, we selected near-infrared spectral data from 238 tobacco blending module samples collected between 2017 and 2019. Combining the power of XGBoost and deep learning, we constructed a flavor prediction model based on feature variables. The XGBoost model was utilized to extract essential information from the high-dimensional near-infrared spectra, while a convolutional neural network with an attention mechanism was employed to predict the flavor type of the modules. The experimental results demonstrate that our model exhibits excellent learning and prediction capabilities, achieving an impressive 95.54% accuracy in flavor category recognition. Therefore, the proposed method of predicting flavor types based on near-infrared spectral features plays a valuable role in facilitating rapid positioning, scientific evaluation, and cigarette formulation design for tobacco blending modules, thereby assisting decision-making processes in the tobacco industry.

Estimating Fine Fuel Load Using Sentinel-2A Imagery and Machine Learning: A Case Study in the Mountainous Forests of Changsha, China

Fine fuel load (FFL) is a crucial variable influencing the occurrence of wildfire. Accurate knowledge of the distribution of FFL in mountainous forests is essential for ongoing wildfire risk management and the stability of mountain ecosystems. Traditional methods of estimating forest fuel load typically involve ground surveys combined with remote sensing, which can be costly and inefficient. Therefore, low-cost, large-scale FFL estimation remains challenging. In this study, Sentinel-2A satellite imagery from the Changsha forest region was used as the data source. Firstly, different feature variables were constructed based on false-color (B843), true-color (B432), four-band (B8432) combinations, and the Normalized Difference Water Index (NDWI). Subsequently, a machine learning approach based on random convolution was employed to estimate FFL. This study also included accuracy assessments of the estimation results and the creation of FFL maps for the study area. The results showed that the FFL estimation based on the B8432 band combination achieved the highest accuracy, with RMSE and R2 values of 5.847 t·hm−2 and 0.656, respectively. FFL estimation results based on false-color imagery followed, with true-color imagery and NDWI index-based estimation results exhibiting lower accuracy. This study offers critical FFL insights using random convolution techniques applied to Sentinel-2A imagery, enhancing the ability to monitor and manage forest fuel conditions effectively, thereby facilitating more informed regional wildfire risk management strategies.

Study on Seasonal Characteristics and Causes of Marine Heatwaves in the South China Sea over Nearly 30 Years

Marine heatwaves (MHWs) are becoming more frequent and intense in many regions around the world, as well as in China’s marginal seas. However, the seasonal characteristics and associated physical drivers of MHWs are largely unknown. In this study, we analyze, based on multiple reanalysis and numerical model data, the seasonal characteristics and causes of MHWs in the South China Sea (SCS) over a near 30-year period (1991–2022). There exist significant seasonal variabilities in the spatiotemporal features and formation mechanisms of MHWs. MHWs in the SCS show significant increasing trends in terms of frequency, duration, and intensity. MHWs during the summer half-year are stronger than the winter half-year as a whole, with them being more likely to occur over the eastern SCS in the summer half-year and the western region in the winter half-year. However, the increasing trend of MHWs in the winter half-year exceed those in the summer. Additionally, we find that MHWs are associated with the unusually strong west Pacific subtropical high (WPSH) both in the summer and winter half-years. Nevertheless, the dominant factors for MHWs are different in the varied seasons. According to upper ocean temperature equation analysis, surface heat flux anomalies (especially shortwave radiation flux) are major effect factors in the summer half-year, while ocean dynamic processes play the main role in the winter half-year. An analysis of the typical MHWs also proves this conclusion. Moreover, MHWs occurring in winter are often accompanied by temperature anomalies within the mixed-layer depth. The findings imply that the formation mechanisms and space–time distribution of MHWs exist with a seasonal contrast in the SCS, rather than simply being due to large-scale circulation and flux anomalies. This may provide a useful reference for a deeper understanding and forecasting of MHWs under different seasons and weather.

Study protocol for “In-vehicle sensors to detect changes in cognition of older drivers”

BackgroundDriving is a complex behavior that may be affected by early changes in the cognition of older individuals. Early changes in driving behavior may include driving more slowly, making fewer and shorter trips, and errors related to inadequate anticipation of situations. Sensor systems installed in older drivers’ vehicles may detect these changes and may generate early warnings of possible changes in cognition.MethodA naturalistic longitudinal design is employed to obtain continuous information on driving behavior that will be compared with the results of extensive cognitive testing conducted every 3 months for 3 years. A driver facing camera, forward facing camera, and telematics unit are installed in the vehicle and data downloaded every 3 months when the cognitive tests are administered.ResultsData processing and analysis will proceed through a series of steps including data normalization, adding information on external factors (weather, traffic conditions), and identifying critical features (variables). Traditional prediction modeling results will be compared with Recurring Neural Network (RNN) approach to produce Driver Behavior Indices (DBIs), and algorithms to classify drivers within age, gender, ethnic group membership, and other potential group characteristics.ConclusionIt is well established that individuals with progressive dementias are eventually unable to drive safely, yet many remain unaware of their cognitive decrements. Current screening and evaluation services can test only a small number of individuals with cognitive concerns, missing many who need to know if they require treatment. Given the increasing number of sensors being installed in passenger vehicles and pick-up trucks and their increasing acceptability, reconfigured in-vehicle sensing systems could provide widespread, low-cost early warnings of cognitive decline to the large number of older drivers on the road in the U.S. The proposed testing and evaluation of a readily and rapidly available, unobtrusive in-vehicle sensing system could provide the first step toward future widespread, low-cost early warnings of cognitive change for this large number of older drivers in the U.S. and elsewhere.

Prediction of Gleason score in prostate cancer patients based on radiomic features of transrectal ultrasound images.

The aim of this study was to develop a model for predicting the Gleason score of patients with prostate cancer based on ultrasound images. Transrectal ultrasound images of 838 prostate cancer patients from The Cancer Imaging Archive database were included in this cross-section study. Data were randomly divided into the training set and testing set (ratio 7:3). A total of 103 radiomic features were extracted from the ultrasound image. Lasso regression was used to select radiomic features. Random forest and broad learning system (BLS) methods were utilized to develop the model. The area under the curve (AUC) was calculated to evaluate the model performance. After the screening, 10 radiomic features were selected. The AUC and accuracy of the radiomic feature variables random forest model in the testing set were 0.727 (95% CI, 0.694-0.760) and 0.646 (95% CI, 0.620-0.673), respectively. When PSA and radiomic feature variables were included in the random forest model, the AUC and accuracy of the model were 0.770 (95% CI, 0.740-0.800) and 0.713 (95% CI, 0.688-0.738), respectively. While the BLS method was utilized to construct the model, the AUC and accuracy of the model were 0.726 (95% CI, 0.693-0.759) and 0.698 (95% CI, 0.673-0.723), respectively. In predictions for different Gleason grades, the highest AUC of 0.847 (95% CI, 0.749-0.945) was found to predict Gleason grade 5 (Gleason score ≥9). A model based on transrectal ultrasound image features showed a good ability to predict Gleason scores in prostate cancer patients. This study used ultrasound-based radiomics to predict the Gleason score of patients with prostate cancer.

10075-CO-3 PREDICTION OF KPS DETERIORATION IN GLIOBLASTOMA USING MULTIMODAL LEARNING WITH PRE- AND POST-OPERATIVE MRI IMAGES AND CLINICAL INFORMATION

Abstract BACKGROUND Previous AI studies on glioblastoma have mostly focused on predicting genetic subtypes or survival outcomes using preoperative imaging alone. We aimed to investigate the feasibility of utilizing postoperative imaging in combination with clinical chart data for multimodal learning, to evaluate systemic functional outcome. METHODS We included 77 glioblastoma cases that received treatment at Kyoto University between December 2001 and January 2022. These cases had both pre- and postoperative MRI images, including T1WI, CE-T1WI, FLAIR, ADC, and DWI. Molecular diagnoses were confirmed for all cases. First, using external data (BRaTS2020), we constructed models for tumor segmentation prediction from MRI images followed by an image feature extraction model using an autoencoder (VAE), extracting 192 image feature variables from preoperative and postoperative images. These variables were combined with 24 clinical variables, and 6 variables of radiologists' findings. We built a CNN model to predict a decrease of 30% or more KPS at 6 months and evaluated its accuracy using the AUC. Feature importance was calculated using PermutationImportance. RESULTS When predicting KPS deterioration using only clinical data, the AUC was 0.79. When combined with radiologists' findings, the AUC increased to 0.84. Combination with preoperative image features resulted in an AUC of 0.78. On the other hand, when combined with both pre- and postoperative image features, the AUC increased to 0.94. Both pre- and post-operative images were identified as important features. DISCUSSION Our findings demonstrate the utility of incorporating postoperative imaging into multimodal learning for predicting systemic functional outcomes in glioblastoma patients. Further validation in independent cases from different medical facilities is warranted to confirm these results.

Application of colorimetric sensor array coupled with chemometric methods for monitoring the freshness of snakehead fillets

Total volatile basic nitrogen content (TVB-N) is an important index of freshness for snakehead. This paper attempted the feasibility of determining TVB-N content level in snakehead fillets by a colorimetric sensor array (CSA) composed of twelve porphyrin materials and eight pH indicators. The nine feature variables in RGB, HSV and CIE L*a*b* color spaces were obtained by differentiating the images of the CSA before and after exposure to the headspace-gas of the samples. Competitive adaptive reweighted sampling combined with partial least squares regression (CARS-PLS) was used to build the relationship between the TVB-N content and the feature variables of CSA, and to select meaningful color-sensitive materials. The results showed that CARS-PLS had a correlation coefficient of 0.9325 in the prediction set and selected 13 informative color-sensitive materials. This study demonstrated that the CSA with CARS-PLS algorithm could be used successfully to quantify and monitor the TVB-N in snakehead fillets.

High-resolution measurements of swordfish skin surface roughness

The three-dimensional morphology of swordfish skin roughness remains poorly understood. Subsequently, its importance to the overall physiology and hydrodynamic performance of the swordfish is yet to be determined. This is at least partly attributable to the inherent difficulty in making the required measurements of these complex biological surfaces. To address this, here two sets of novel high-resolution measurements of swordfish skin, obtained using a modular optical coherence tomography system and a gel-based stereo-profilometer, are reported and compared. Both techniques are shown to provide three-dimensional morphological data at micron-scale resolution. The results indicate that the skin surface is populated with spiny roughness elements, typically elongated in the streamwise direction, in groups of up to six, and in good agreement with previously reported information based on coarser measurements. In addition, our data also provide new information on the spatial distribution and variability of these roughness features. Two approaches, one continuous and another discrete, are used to derive various topographical metrics that characterize the surface texture of the skin. The information provided here can be used to develop statistically representative synthetic models of swordfish skin roughness.

Predicting the risk of stroke based on machine learning

Stroke, the second largest leading cause of death among all chronic diseases, is affecting about 101 million people in the world currently. It is estimated that this number of stroke cases will increase by 2.25 by the year 2050. Considering the large number of potential patients with stroke, a mathematical model is designed to predict ones risk of having a stroke in the future based on ones basic health data using machinery methods. Using the algorithm of Logistic Regression, the model reaches an accuracy of 92.28% when predicting whether one has a stroke, the model also validates that hypertension is the leading cause of the incidence of stroke by finding out the highest correlation value among all the feature variables. People who would like to know their probability of having a stroke can use the model, then they can have some precautionary measures to lower the likelihood of happening of stroke based on the prediction given, which helps save the medical costs and overuse of medical resources. Governments can enact policies and allocate medical resources based on the predictions made by the model.

A resource for development and comparison of multimodal brain 3 T MRI harmonisation approaches

Abstract Despite the huge potential of magnetic resonance imaging (MRI) in mapping and exploring the brain, MRI measures can often be limited in their consistency, reproducibility, and accuracy which subsequently restricts their quantifiability. Nuisance nonbiological factors, such as hardware, software, calibration differences between scanners, and post-processing options, can contribute to, or drive trends in, neuroimaging features to an extent that interferes with biological variability. Such lack of consistency, known as lack of harmonisation, across neuroimaging datasets poses a great challenge for our capabilities in quantitative MRI. Here, we build a new resource for comprehensively mapping the extent of the problem and objectively evaluating neuroimaging harmonisation approaches. We use a travelling-heads paradigm consisting of multimodal MRI data of 10 travelling subjects, each scanned at five different sites on six different 3 T scanners from all the three major vendors and using five neuroimaging modalities, providing more comprehensive coverage than before. We also acquire multiple within-scanner repeats for a subset of subjects, setting baselines for multimodal scan-rescan variability. Having extracted hundreds of imaging-derived phenotypes, we compare three forms of variability: (i) between-scanner, (ii) within-scanner (within-subject), and (iii) biological (between-subject). We characterise the reliability of features across scanners and use our resource as a testbed to enable new investigations that until now have been relatively unexplored. Specifically, we identify optimal pipeline processing steps that minimise between-scanner variability in extracted features (implicit harmonisation). We also test the performance of post-processing harmonisation tools (explicit harmonisation) and specifically check their efficiency in reducing between-scanner variability against baseline standards provided by our data. Our explorations allow us to come up with good practice suggestions on processing steps and sets of features where results are more consistent, while our publicly released dataset (which we refer to as ON-Harmony) establishes references for future studies in this field.

Evaluating Feature Selection Methods and Machine Learning Algorithms for Mapping Mangrove Forests Using Optical and Synthetic Aperture Radar Data

Mangrove forests, mostly found in the intertidal zone, are among the highest-productivity ecosystems and have great ecological and economic value. The accurate mapping of mangrove forests is essential for the scientific management and restoration of mangrove ecosystems. However, it is still challenging to perform the rapid and accurate information mapping of mangrove forests due to the complexity of mangrove forests themselves and their environments. Utilizing multi-source remote sensing data is an effective approach to address this challenge. Feature extraction and selection, as well as the selection of classification models, are crucial for accurate mangrove mapping using multi-source remote sensing data. This study constructs multi-source feature sets based on optical (Sentinel-2) and SAR (synthetic aperture radar) (C-band: Sentinel-1; L-band: ALOS-2) remote sensing data, aiming to compare the impact of three feature selection methods (RFS, random forest; ERT, extremely randomized tree; MIC, maximal information coefficient) and four machine learning algorithms (DT, decision tree; RF, random forest; XGBoost, extreme gradient boosting; LightGBM, light gradient-boosting machine) on classification accuracy, identify sensitive feature variables that contribute to mangrove mapping, and formulate a classification framework for accurately recognizing mangrove forests. The experimental results demonstrated that using the feature combination selected via the ERT method could obtain higher accuracy with fewer features compared to other methods. Among the feature combinations, the visible bands, shortwave infrared bands, and the vegetation indices constructed from these bands contributed the greatest to the classification accuracy. The classification performance of optical data was significantly better than SAR data in terms of data sources. The combination of optical and SAR data could improve the accuracy of mangrove mapping to a certain extent (0.33% to 4.67%), which is essential for the research of mangrove mapping in a larger area. The XGBoost classification model performed optimally in mangrove mapping, with the highest overall accuracy of 95.00% among all the classification models. The results of the study show that combining optical and SAR remote sensing data with the ERT feature selection method and XGBoost classification model has great potential for accurate mangrove mapping at a regional scale, which is important for mangrove restoration and protection and provides a reliable database for mangrove scientific management.

Dynamic nonlinear soft sensor modelling method using linear slow feature analysis and least squares support vector regression for batch processes

AbstractData‐driven soft sensor models have been extensively utilized in industrial processes. Batch processes are usually employed to manufacture low‐volume and high value‐added products in chemical, materials, and pharmaceutical industries. The most distinctive features of batch process lie in nonlinear, repetition, and slow time varying characteristics. In this paper, a data‐driven soft sensor modelling method based on linear slow feature analysis (LSFA) and least squares support vector regression (LSSVR) is proposed. In this method, LSFA was used to effectively capture the driving force behind the data features that change as slowly as possible. Then, a LSSVR model was constructed between the extracted slow feature variables and quality variables. Finally, a numerical example, industrial penicillin fermentation processes, and cobalt oxalate synthesis process were utilized to confirm the prediction accuracy and model reliability of the proposed approach.

Reproducible Python workflow for multi‐site resting‐state EEG analysis: From raw data to group classification

AbstractBackgroundAmong promising markers for neurodegenerative disorders, electroencephalogram (EEG) represents a non‐invasive and potentially portable alternative. Standardized methods for preprocessing may contribute to less inter‐observer variability, while pooling multi‐centric EEG data may provide more generalizable results. We aim to formulate a reproducible pipeline for group‐level classification (of Parkinson’s Disease ‐PD‐) based on various features of resting‐state EEG recordings (rsEEG) collected in multi‐site studies.MethodsWe used four datasets (acquired in Colombia, Finland, and the USA) consisting of 169 subjects (84 PD; 85 non‐PD healthy controls). Our workflow included rsEEG raw files standardization, preprocessing, spectral bandpowers and multiple entropy features extraction on sensor space, harmonization of multi‐site features, and group‐level multiclass classification using supervised machine learning methods. Center/scanner‐related variability effects in the extracted features were harmonized using the ComBat approach and its variants 1. Finally, sequential feature selection (SFS) and binary classification (Non‐PD vs. PD) was conducted using XGradientBoosting 2. The full workflow is detailed in Figure 1.Resultst‐distributed stochastic neighbor embedded (tSNE) plots showed that center‐related variability in spectral features was reduced after harmonization using the ComBat method and its bootstrapped variant, Figure 2. SFS of rsEEG features was performed on 224 harmonized features, showing the highest performance when alpha, pre‐alpha, and slow‐theta bands were considered for subsequent classification. The XGradientBoosting classifier showed an AUC of 0.826 in separating PD from non‐PD, Figure 3.ConclusionsIn multicenter studies, the harmonization of rsEEG features at the sensor space using ComBat can contribute to controlling batch effects. XGradientBoosting with rsEEG features and SFS performed well in classifying non‐PD vs. PD.Intended to be a practical tool for group‐level analysis of neurodegenerative disorders based on sensor space rsEEG features, our workflow is openly available at (https://github.com/alberto‐jj/raw_to_classification) and can be easily adapted to multi‐class classification and regression tasks.

Applying Artificial Intelligence Machine Learning to Discover Factors That Shorten Duration of Treatment in Patients with Multiple Myeloma Using Real-World Patient-Level Longitudinal Data from Medical Data Vision Database

Introduction: To maximize the benefit of treatment in multiple myeloma (MM), it is important to continue treatment for appropriate duration . Our recent real-world study using the Medical Data Vision (MDV) claim database showed that median duration of treatment (DoT) of 1 line (L), 2L and 3L treatment was rather short (range: 6.7 to 8.6 months) in Japanese non-transplant patients with MM (Handa H, et al. Plos one. 2023). However, the broad range of treatment patterns in Japan makes it difficult to identify the cause of shortened DoT. Therefore, we assessed the predictive factors that shorten DoT at threshold of <3, <6 and <12 months by applying machine learning (ML) model using real-world clinical data of MM patients from the MDV database. Methods: This retrospective cohort study was conducted on 4848 individual treatments (2,762 patients; 1L for stem cell transplant-not conducted, 2L/3L) in patients (aged ≥18 years) with transplant-ineligible (newly diagnosed or relapsed/refractory) MM between 2003-21. An explainable deep learning model (point-wise linear model, PWL) was developed using 647 feature variables (evaluated before, during, and after treatment; Table) from the MDV database to identify the important predictors associated with DoT. The PWL model generated a custom-made logistic regression model for each sample, which was very familiar to medical researchers and allowed to analyze the importance of the feature variables. Predictive performance of the PWL model was compared with elastic-net regularized logistic regression and XGBoost models and calculated by area under the curve (AUC) and evaluated by 10-fold double cross validation. To stratify the patients with shorter DoT, a clustering analysis of 4848 individual treatment was conducted based on the similarity of coefficients of the custom-made logistic regression models generated from the PWL model. To understand characteristics of samples belonging to each cluster, the relationship between predicted DoT and MM treatment-related comorbidity management was investigated ( Table). Results: The AUC score of the PWL model to predict DoT of <3, <6 and <12 months was 0.61, 0.64 and 0.66, respectively ( Figure). As per clustering analysis, there were 2 clusters (cluster A and B) at DoT of <3 months and 3 clusters (cluster A, B and C) each at DoT of <6 and <12 months, respectively. At <3 months, cluster B showed lower median of prediction probability, and had a significantly (P<0.01) higher previous treatment lines, pre-treatment Charlson Comorbidity Index (CCI) versus cluster A. At <6 months, cluster C showed lower median of prediction probability, and had significantly (P<0.01) higher CCI and comorbidities, and cluster B had significantly (P<0.01) lower estimated glomerular filtration rate, among the clusters. At <12 months, cluster C showed lower median of prediction probability, and had significantly (P<0.01) higher lactate dehydrogenase (LD), previous treatment lines, pre-treatment CCI, degree of care requirement at diagnosis of MM, pre-treatment tube and parenteral nutrition, and comorbidities, among clusters. Additionally, in cluster B (at <3 months), and cluster C (at <6 and <12 months), use of immunomodulatory drugs (IMiDs) in treatment for MM was significantly higher in patients who met predicted DoT at each threshold versus the ones who were not. However, age and history of infectious diseases (antibacterial agents, γ globulin, etc.) was not affected. In addition, IMiDs and aspirin use at <3 and <6 months, and IMiDs use at <12 months had significant differences in events during and after treatment, which are important for considering treatment strategies for MM patients and may lead to the optimization of treatment plans. Study limitation included exclusion of important prognostic factors (cytogenetic profile, R-ISS stage, and presence of all plasmacytomas). Conclusions: TheML technique with PWL model yielded efficient results to understand the trend associated with event in treatment and characteristics of Japanese patients with MM whose DoT were shortened. Patient's condition and management related factors might have associated with shortened DoT, in consistent with clinical findings. These results are critical to continually improving MM management, achieving optimal patients care, and understanding the burden of disease and health services available to Japanese patients with MM in the real-world.