Recursive Feature Elimination Method Research Articles

Design of smoke detection system using deep learning and sensor fusion with recursive feature elimination cross-validation

The fire safety system is an important component that controls material and immaterial losses. Fire disasters are generally indicated by the appearance of excess smoke and changes in temperature, pressure, and changes in other parameters in the environment. Conventional smoke sensors are limited in reading parameter changes around their environment, making them less effective in early fire detection. This study aims to design a smoke detection system as an early fire detection system, using sensor fusion based on deep learning using the recursive feature elimination method with cross-validation (RFECV) using a random forest classifier used to select optimal parameters from public datasets as the basis for determining the sensor to be used. Based on the RFECV optimal feature, a deep learning algorithm was performed and obtained an accuracy of 0.99, a precision of 0.99, a recall of 1.00, and an F1 score of 0.99, with a latency time of 34.02 μs, which is 71.76% times faster than the original model.

Developing a Model to Predict Self-Reported Student Performance during Online Education Based on the Acoustic Environment

In recent years, great developments in online university education have been observed, favored by advances in ICT. There are numerous studies on the perception of academic performance in online classes, influenced by aspects of a very diverse nature; however, the acoustic environment of students at home, which can certainly affect the performance of academic activities, has barely been evaluated. This study assesses the influence of the home acoustic environment on students’ self-reported academic performance. This assessment is performed by calculating prediction models using the Recursive Feature Elimination method with 40 initial features and the following classifiers: Random Forest, Gradient Boosting, and Support Vector Machine. The optimal number of predictors and their relative importance were also evaluated. The performance of the models was assessed by metrics such as the accuracy and the area under the receiver operating characteristic curve (ROC_AUC-score). The model with the smallest optimal number of features (with 14 predictors, 9 of them about the perceived acoustic environment) and the best performance achieves an accuracy of 0.7794; furthermore, the maximum difference for the same algorithm between using 33 and 14 predictors is 0.03. Consequently, for simplicity and the ease of interpretation, models with a reduced number of variables are preferred.

Diagnostic significance of multisequence MRI radiomics models in distinguishing benign and malignant spinal fractures

PurposeTo investigate the diagnostic value of multi-sequence magnetic resonance imaging (MRI) in benign and malignant spinal fractures. Materials and methodsMRI data of patients with pathologically confirmed malignant vertebral fractures were retrospectively collected and compared with those with benign vertebral fractures. The image omics features of T1-weighted imaging (T1WI) and T2-weighted imaging (T2WI) were extracted respectively, and were randomly divided into training and test sets at 8:2. The training set adopted recursive feature elimination method and minimum absolute contraction and selection operator (LASSO) regression to screen the variables. Three image omics models, T1WI, T2WI and the combination of two sequences, were constructed by logistic regression, and the diagnostic efficiency of each model was verified by the test set. The receiver operating curve evaluated the diagnostic efficiency of the model. ResultsA total of 111 vertebrae with fracture were included in 97 patients, 55 with benign fracture and 56 with malignant fracture. In the training set, the AUC, sensitivity and specificity of the T1WI model were 0.892, 88.9% and 81.4%, respectively. In T2WI model, the results were 0.924, 82.2% and 95.3%, respectively. The combined models of the two sequences were 0.934, 90.0% and 87.5%, respectively. In the test set, the AUC, sensitivity and specificity of T1WI model were 0.877, 80.0% and 92.3%, respectively. In T2WI model, the results were 0.923, 90.0% and 92.3%, respectively. The combined models were 0.933, 100.0% and 87.5%, respectively. The diagnostic efficiency of the two-sequence combined image omics model was better than that of the radiologist. ConclusionThe combined model, incorporating T1WI and T2WI, outperformed single-sequence models in terms of diagnostic accuracy. The integration of multisequence radiomic features provided enhanced texture information, contributing to superior accuracy and efficiency in diagnosing malignant spinal fractures.

Combining spectrum, thermal, and texture features using machine learning algorithms for wheat nitrogen nutrient index estimation and model transferability analysis

The nitrogen nutrition index (NNI) has been extensively applied for the diagnosis of crop nitrogen status, providing insights into efficient nitrogen utilization and plant growth. In this study, we utilized a low-altitude unmanned aerial vehicle (UAV) platform, equipped with multispectral (MS), red–green–blue (RGB), and thermal infrared (TIR) cameras, to comprehensively capture wheat spectral information. The analysis of the relationship between NNI and relative yield revealed an initially linear relationship, which saturated for high NNI values. To enhance accuracy and minimize complexity, we employed a random forest (RF) – recursive feature elimination (RFE) method to select features as inputs for four machine learning (ML) models: back propagation neural network (BPNN), extreme learning machine (ELM), support vector regression (SVR), and Gaussian process regression (GPR). After feature selection, the prediction accuracies of single-sensor models were ranked as: MS > RGB > TIR. The R2 values for the four ML models were in the range of 0.54–0.75. Among multi-sensor combinations, the GPR with MS + RGB + TIR input features achieved the best results with R2 = 0.89 and RPD = 2.52. Further, the dataset was partitioned into six subsets based on location and cultivar variety to evaluate model transferability. The results showed that the transferability largely suffered during the bivariate conditions of different varieties at different locations; the transferability of the model was average improved by 11 % when GPR was combined with transfer component analysis (TCA). The accuracy and transferability of the NNI estimation models significantly improved, offering valuable guidance and methodological support for diagnosing the nitrogen nutrient status of wheat.

Aquatic vegetation mapping with UAS-cameras considering phenotypes

Aquatic vegetation species at the genus level in an oxbow lake were identified in Hungary based on a multispectral Uncrewed Aerial System (UAS) survey within an elongated oxbow lake area of the Tisza River under continental climate. Seven and 13 classes were discriminated using three different classification methods (Support Vector Machine [SVM], Random Forest [RF], and Multivariate Adaptive Regression Splines [MARS]) using different input data in ten combinations: original spectral bands, spectral indices, Digital Surface Model (DSM), and Haralick texture indices. We achieved a high (97.1%) overall accuracies (OAs) by applying the SVM classifier, but the RF performed only <1% worse, as it was represented in the first places of the classification rank before the MARS. The highest classification accuracies (>84% OA) were obtained using the most important variables derived by the Recursive Feature Elimination (RFE) method. The best classification required DSM as an input variable. The poorest classification performance belonged to the model that used only texture indices or spectral indices. On the class level, Stratiotes aloides exhibit the lowest degree of separability compared to the other classes. Accordingly, we recommend using supplementary input data for the classifications besides the original spectral bands, for example, DSM, spectral, and texture indices, as these variables significantly improve the classification accuracies in the proper combinations of the input variables.

A hierarchical recursive feature elimination algorithm to develop brain computer interface application of user behavior for statistical reasoning and decision making

BackgroundWith the aid of a brain computer interface (BCI), users can communicate and receive signals wirelessly or over wired connections to operate smart devices. A BCI classifier's architecture is quite difficult since numerous elements should be combined. These elements are made up of brain signals, which also include high levels of weak sounds that could provide reliable participant recordings of daily activities. We must use computer vision techniques to create a model in order to control those information. The high dimension and volume of signals present the classification classifier with its primary obstacles. New methodDue to this, we extracted and classified the brain activity in this study, and we also presented a novel hierarchical recursive feature elimination method that we refer to as HRFE embracing noisy additions. HRFE makes a variety of categorization suggestions to eliminate bias in classifying BCI systems of different types. We put the HRFE to the test on two BCI signal data sets—specifically, dataset I and BCI contests III—using shallow and deep convolution network classification techniques. Just a grid of assets is used to create electrocorticography (ECoG) signals on the contralateral (right) motor cortex, and these signals are recorded in the BCI contests III database. ResultsUsing ECoG signals, we choose the top 20 features that have the biggest effects on distortion and classification selection. Comparison with existing methodsThe simulation findings show that HRFE has a significant computer vision enhancement when compared to comparable feature selection methods in the literature, particularly for ECoG signal, which achieves about 93% reliability.

Computed tomography-based radiomics to predict early recurrence of hepatocellular carcinoma post-hepatectomy in patients background on cirrhosis.

The prognosis for hepatocellular carcinoma (HCC) in the presence of cirrhosis is unfavourable, primarily attributable to the high incidence of recurrence. To develop a machine learning model for predicting early recurrence (ER) of post-hepatectomy HCC in patients with cirrhosis and to stratify patients' overall survival (OS) based on the predicted risk of recurrence. In this retrospective study, 214 HCC patients with cirrhosis who underwent curative hepatectomy were examined. Radiomics feature selection was conducted using the least absolute shrinkage and selection operator and recursive feature elimination methods. Clinical-radiologic features were selected through univariate and multivariate logistic regression analyses. Five machine learning methods were used for model comparison, aiming to identify the optimal model. The model's performance was evaluated using the receiver operating characteristic curve [area under the curve (AUC)], calibration, and decision curve analysis. Additionally, the Kaplan-Meier (K-M) curve was used to evaluate the stratification effect of the model on patient OS. Within this study, the most effective predictive performance for ER of post-hepatectomy HCC in the background of cirrhosis was demonstrated by a model that integrated radiomics features and clinical-radiologic features. In the training cohort, this model attained an AUC of 0.844, while in the validation cohort, it achieved a value of 0.790. The K-M curves illustrated that the combined model not only facilitated risk stratification but also exhibited significant discriminatory ability concerning patients' OS. The combined model, integrating both radiomics and clinical-radiologic characteristics, exhibited excellent performance in HCC with cirrhosis. The K-M curves assessing OS revealed statistically significant differences.

Hybrid forecasting model of building cooling load based on combined neural network

The inefficient operation of air-conditioning system, which leads to its high-energy consumption, has aroused far-ranging concern at present. And an accurate prediction of building cooling load can be conducive to matching the supply and demand of the air-conditioning system, thus improving its operational efficiency. Therefore, a hybrid forecasting model (BAS-GRNN&LSTM) combing generalized regression neural network (BAS-GRNN) and long short-term memory neural network (BAS-LSTM) optimized by beetle antennae search algorithm is proposed for building cooling load prediction. First, the factors affecting the cooling load are comprehensively analyzed, and the random forest combined with recursive feature elimination (RF-RFE) method is used for feature selection. Then, the BAS-GRNN&LSTM model is developed to forecast. Finally, a simulation experiment is carried out using the measured data of a large building in north of China. Compared with GRNN, LSTM, BAS-GRNN and BAS-LSTM, BAS-GRNN&LSTM performs better in five performance evaluation metrics, RMSE, MAPE, RRE, MBE and R2, demonstrating its higher prediction accuracy. Furthermore, the detailed algorithm performance analysis shows that the proposed hybrid model has significant advantages in robustness, generalization capability and computational complexity, and can be well applied to building cooling load prediction and beneficial to the optimal control of building air-conditioning system.

Interpretable Machine Learning Framework to Predict the Glass Transition Temperature of Polymers.

The glass transition temperature of polymers is a key parameter in meeting the application requirements for energy absorption. Previous studies have provided some data from slow, expensive trial-and-error procedures. By recognizing these data, machine learning algorithms are able to extract valuable knowledge and disclose essential insights. In this study, a dataset of 7174 samples was utilized. The polymers were numerically represented using two methods: Morgan fingerprint and molecular descriptor. During preprocessing, the dataset was scaled using a standard scaler technique. We removed the features with small variance from the dataset and used the Pearson correlation technique to exclude the features that were highly connected. Then, the most significant features were selected using the recursive feature elimination method. Nine machine learning techniques were employed to predict the glass transition temperature and tune their hyperparameters. The models were compared using the performance metrics of mean absolute error (MAE), root mean square error (RMSE), and coefficient of determination (R2). We observed that the extra tree regressor provided the best results. Significant features were also identified using statistical machine learning methods. The SHAP method was also employed to demonstrate the influence of each feature on the model's output. This framework can be adaptable to other properties at a low computational expense.

Comprehensive evaluation and performance analysis of machine learning in heart disease prediction

Heart disease is a leading cause of mortality on a global scale. Accurately predicting cardiovascular disease poses a significant challenge within clinical data analysis. The present study introduces a prediction model that utilizes various combinations of information and employs multiple established classification approaches. The proposed technique combines the genetic algorithm (GA) and the recursive feature elimination method (RFEM) to select relevant features, thus enhancing the model’s robustness. Techniques like the under sampling clustering oversampling method (USCOM) address the issue of data imbalance, thereby improving the model’s predictive capabilities. The classification challenge employs a multilayer deep convolutional neural network (MLDCNN), trained using the adaptive elephant herd optimization method (AEHOM). The proposed machine learning-based heart disease prediction method (ML-HDPM) demonstrates outstanding performance across various crucial evaluation parameters, as indicated by its comprehensive assessment. During the training process, the ML-HDPM model exhibits a high level of performance, achieving an accuracy rate of 95.5% and a precision rate of 94.8%. The system’s sensitivity (recall) performs with a high accuracy rate of 96.2%, while the F-score highlights its well-balanced performance, measuring 91.5%. It is worth noting that the specificity of ML-HDPM is recorded at a remarkable 89.7%. The findings underscore the potential of ML-HDPM to transform the prediction of heart disease and aid healthcare practitioners in providing precise diagnoses, exerting a substantial influence on patient care outcomes.

Feature selection using non-parametric correlations and important features on recursive feature elimination for stock price prediction

Stock price prediction using machine learning is a rapidly growing area of research. However, the large number of features that can be used can complicate the learning process. The feature selection method that can be used to overcome this problem is recursive feature elimination. Standard recursive feature elimination carries the risk of producing inaccurate algorithms because the top-ranked features are not necessarily the most important features. This research proposes a feature selection method that combines important features and nonparametric correlation in recursive feature elimination for stock price prediction. The data features used are technical indicators and stock price history. The recursive feature elimination method is modified with important features and nonparametric correlation features. The strategy for combining important features and non-parametric features is average weight, 25:75% weight, 75:25% weight, maximum weight, and minimum weight. The performance evaluation results show that the proposed feature selection method succeeded in obtaining small error values. The proposed method for predicting PT Bank Rakyat Indonesia Tbk (BBRI) stock prices obtains mean squared error, root mean square error, mean absolute error, and mean absolute percentage error evaluation values of 0.0000336, 0.00577, 0.00459, and 1.78%, respectively. This shows that recursive feature elimination with feature selection that combines important features and non-parametric correlation works better than the original recursive feature elimination at predicting stock prices.

Estimates of Crop Yield Anomalies for 2022 in Ukraine Based on Copernicus Sentinel-1, Sentinel-3 Satellite Data, and ERA-5 Agrometeorological Indicators.

The study explores the feasibility of adapting the EOStat crop monitoring system, originally designed for monitoring crop growth conditions in Poland, to fulfill the requirements of a similar system in Ukraine. The system utilizes satellite data and agrometeorological information provided by the Copernicus program, which offers these resources free of charge. To predict crop yields, the system uses several factors, such as vegetation condition indices obtained from Sentinel-3 Ocean and Land Color Instrument (OLCI) optical and Sea and Land Surface Temperature Radiometer (SLSTR). It also incorporates climate information, including air temperature, total precipitation, surface radiation, and soil moisture. To identify the best predictors for each administrative unit, the study utilizes a recursive feature elimination method and employs the Extreme Gradient Boosting regressor, a machine learning algorithm, to forecast crop yields. The analysis indicates a noticeable decrease in crop losses in 2022 in certain regions of Ukraine, compared to the previous year (2021) and the 5-year average (2017-2021), specifically for winter crops and maize. Considering the reduction in yield, it is estimated that the decline in production of winter crops in 2022 was up to 20%, while for maize, it was up to 50% compared to the decline in production.

An Analysis of Machine Learning for Detecting Depression, Anxiety, and Stress of Recovered COVID-19 Patients

Objectives: This study explores different machine learning models (KNN: k-nearest neighbor, MLP: Multilayer Perceptron, SVM: Support Vector Machine) to identify the optimal model for accurate and rapid mental health detection among the recovered COVID-19 patients. Other techniques are also investigated, such as feature selection (Recursive Feature Elimination (RFE) and Extra Trees (ET) methods) and hyper-parameter tuning, to achieve a system that could effectively and quickly indicate mental health. Method/Analysis: To achieve the objectives, the study employs a dataset collected from recovered COVID-19 patients, encompassing information related to depression, anxiety, and stress. Machine learning models are utilized in the analysis. Additionally, feature selection methods and hyper-parameter tuning techniques are explored to enhance the model’s predictive capabilities. The performance of each model is assessed based on accuracy metrics. Findings: The experimental results show that SVM is the most suitable model for accurately predicting an individual’s mental health among recovered COVID-19 patients (accuracy ≥ 0.984). Furthermore, the ET method is more effective than the RFE method for feature selection in the anxiety and stress datasets. Novelty/Improvement:The study lies in the understanding of predictive modeling for mental health and provides insights into the choice of models and techniques for accurate and early detection. Doi: 10.28991/HEF-2024-05-01-01 Full Text: PDF

Integrating Meteorological and Remote Sensing Data to Simulate Cropland Nocturnal Evapotranspiration Using Machine Learning

Evapotranspiration (ET) represents a significant component of the global water flux cycle, yet nocturnal evapotranspiration (ETn) is often neglected, leading to underestimation of global evapotranspiration. As for cropland, accurate modeling of ETn is essential for rational water management and is important for sustainable agriculture development. We used random forest (RF) to simulate ETn at 16 globally distributed cropland eddy covariance flux sites along with remote sensing and meteorological factors. The recursive feature elimination method was used to remove unimportant variables. We also simulated the ETn of C3 and C4 crops separately. The trained RF resulted in a determination coefficient (R2) (root mean square error (RMSE)) of 0.82 (7.30 W m−2) on the testing dataset. C3 and C4 crops on the testing dataset resulted in an R2 (RMSE) of 0.86 (5.59 W m−2) and 0.55 (4.86 W m−2) for the two types of crops. We also showed that net radiation is the dominant factor in regulating ETn, followed by 2 m horizontal wind speed and vapor pressure deficit (VPD), and these three meteorological factors showed a significant positive correlation with ETn. This research demonstrates that RF can simulate ETn from crops economically and accurately, providing a methodological basis for improving global ETn simulations.

Bioinformatics Analysis and Experimental Validation of Mitochondrial Autophagy Genes in Knee Osteoarthritis.

Mitochondrial autophagy is closely related to the pathogenesis of osteoarthritis, In order to explore the role of mitochondrial autophagy related genes in knee osteoarthritis (KOA) and its molecular mechanism. KOA-related transcriptome data were extracted from the Gene Expression Omnibus (GEO) database. Differentially expressed mitochondrial autophagy gene (DEMGs) were screened in patients with KOA by differential expression analysis. The STRING website was used to construct a protein-protein interaction (PPI) network among DEMGs. Molecular complex detection (MCODE) method in Cytoscape software was performed to identify hub DEMGs. Support vector machine recursive feature elimination (SVM-RFE) method was used to construct the hub DEMG diagnosis model. Genes with diagnostic value were identified as biomarkers by plotting receiver operating characteristic (ROC) curves and Expression validation. CIBERSORT algorithm was used to calculate the proportion of 22 immune cells in each sample in the GSE114007 dataset. Finally, biomarker expression was verified by qPCR. A total of 15 DEMGs were obtained and enrichment analyses showed that these DEMG strains were mainly enriched in the mitophagy-animal, shigellosis, autophagy-animal and FoxO signal pathways. The PPI network unveiled 13 DEMGs with interactions. In addition, 8 hub DEMGs (ULK1, CALCOCO2, MAP1LC3B, BNIP3L, GABARAPL1, BNIP3, FKBP8 and FOXO3) were obtained for KOA. And 5 model DEMGs (BNIP3L, BNIP3, MAP1LC3B, ULK1 and FOXO3) were screened. The ROC curves revealed that BNIP3 and FOXO3 has strong diagnostic value in these models of DEMG. Immune-infiltration and correlation analysis showed that BNIP3 and FOXO3 were significantly correlated with three different immune cells, including primary B cells, M0 macrophage and M2 macrophage. The cartilage tissue samples qPCR verification results show that FOXO3 and BNIP3 were all down-regulated in KOA (p < 0.01), and the validation results are consistent with the above analysis. BNIP3 and FOXO3 have been identified as biomarkers for the diagnosis of KOA, which might supply a new insight for the pathogenesis and treatment of KOA.

Application of radiomics model based on ultrasound image features in the prediction of carpal tunnel syndrome severity.

The aim of our study is to develop and validate a radiomics model based on ultrasound image features for predicting carpal tunnel syndrome (CTS) severity. This retrospective study included 237 CTS hands (106 for mild symptom, 68 for moderate symptom and 63 for severe symptom). There were no statistically significant differences among the three groups in terms of age, gender, race, etc. The data set was randomly divided into a training set and a test set in a ratio of 7:3. Firstly, a senior musculoskeletal ultrasound expert measures the cross-sectional area of median nerve (MN) at the scaphoid-pisiform level. Subsequently, a recursive feature elimination (RFE) method was used to identify the most discriminative radiomic features of each MN at the entrance of the carpal tunnel. Eventually, a random forest model was employed to classify the selected features for prediction. To evaluate the performance of the model, the confusion matrix, receiver operating characteristic (ROC) curves, and F1 values were calculated and plotted correspondingly. The prediction capability of the radiomics model was significantly better than that of ultrasound measurements when 10 robust features were selected. The training set performed perfect classification with 100% accuracy for all participants, while the testing set performed accurate classification of severity for 76.39% of participants with F1 values of 80.00, 63.40, and 84.80 for predicting mild, moderate, and severe CTS, respectively. Comparably, the F1 values for mild, moderate, and severe CTS predicted based on the MN cross-sectional area were 76.46, 57.78, and 64.00, respectively.. This radiomics model based on ultrasound images has certain value in distinguishing the severity of CTS, and was slightly superior to using only MN cross-sectional area for judgment. Although its diagnostic efficacy was still inferior to that of neuroelectrophysiology. However, this method was non-invasive and did not require additional costs, and could provide additional information for clinical physicians to develop diagnosis and treatment plans.

Prediction of the Fatal Acute Complications of Myocardial Infarction via Machine Learning Algorithms.

Myocardial infarction (MI) is a major cause of death, particularly during the first year. The avoidance of potentially fatal outcomes requires expeditious preventative steps. Machine learning (ML) is a subfield of artificial intelligence science that detects the underlying patterns of available big data for modeling them. This study aimed to establish an ML model with numerous features to predict the fatal complications of MI during the first 72 hours of hospital admission. We applied an MI complications database that contains the demographic and clinical records of patients during the 3 days of admission based on 2 output classes: dead due to the known complications of MI and alive. We utilized the recursive feature elimination (RFE) method to apply feature selection. Thus, after applying this method, we reduced the number of features to 50. The performance of 4 common ML classifier algorithms, namely logistic regression, support vector machine, random forest, and extreme gradient boosting (XGBoost), was evaluated using 8 classification metrics (sensitivity, specificity, precision, false-positive rate, false-negative rate, accuracy, F1-score, and AUC). In this study of 1699 patients with confirmed MI, 15.94% experienced fatal complications, and the rest remained alive. The XGBoost model achieved more desirable results based on the accuracy and F1-score metrics and distinguished patients with fatal complications from surviving ones (AUC=78.65%, sensitivity=94.35%, accuracy=91.47%, and F1-score=95.14%). Cardiogenic shock was the most significant feature influencing the prediction of the XGBoost algorithm. XGBoost algorithms can be a promising model for predicting fatal complications following MI.

A Wearable Inertial Sensor Approach for Locomotion and Localization Recognition on Physical Activity.

Advancements in sensing technology have expanded the capabilities of both wearable devices and smartphones, which are now commonly equipped with inertial sensors such as accelerometers and gyroscopes. Initially, these sensors were used for device feature advancement, but now, they can be used for a variety of applications. Human activity recognition (HAR) is an interesting research area that can be used for many applications like health monitoring, sports, fitness, medical purposes, etc. In this research, we designed an advanced system that recognizes different human locomotion and localization activities. The data were collected from raw sensors that contain noise. In the first step, we detail our noise removal process, which employs a Chebyshev type 1 filter to clean the raw sensor data, and then the signal is segmented by utilizing Hamming windows. After that, features were extracted for different sensors. To select the best feature for the system, the recursive feature elimination method was used. We then used SMOTE data augmentation techniques to solve the imbalanced nature of the Extrasensory dataset. Finally, the augmented and balanced data were sent to a long short-term memory (LSTM) deep learning classifier for classification. The datasets used in this research were Real-World Har, Real-Life Har, and Extrasensory. The presented system achieved 89% for Real-Life Har, 85% for Real-World Har, and 95% for the Extrasensory dataset. The proposed system outperforms the available state-of-the-art methods.

Multisequence MRI-based radiomics signature as potential biomarkers for differentiating KRAS mutations in non-small cell lung cancer with brain metastases

BackgroundKirsten rat sarcoma virus (KRAS) has evolved from a genotype with predictive value to a therapeutic target recently. The study aimed to establish non-invasive radiomics models based on MRI to discriminate KRAS from epidermal growth factor receptor (EGFR) or anaplastic lymphoma kinase (ALK) mutations in lung cancer patients with brain metastases (BM), then further explore the optimal sequence for prediction. MethodsThis retrospective study involved 317 patients (218 patients in training cohort and 99 patients in testing cohort) who had confirmed of KRAS, EGFR or ALK mutations. Radiomics features were separately extracted from T2WI, T2 fluid-attenuated inversion recovery (T2-FLAIR), diffusion weighted imaging (DWI) and contrast-enhanced T1-weighted imaging (T1-CE) sequences. The maximal information coefficient and recursive feature elimination method were used to select informative features. Then we built four radiomics models for differentiating KRAS from EGFR or ALK using random forest classifier. ROC curves were used to validate the capability of the models. ResultsThe four radiomics models for discriminating KRAS from EGFR all worked well, especially DWI and T2WI models (AUCs: 0.942, 0.942 in training cohort, 0.949, 0.954 in testing cohort). When KRAS compared to ALK, DWI and T2-FLAIR models showed excellent performance in two cohorts (AUCs: 0.947, 0.917 in training cohort, 0.850, 0.824 in testing cohort). ConclusionsRadiomics classifiers integrating MRI have potential to discriminate KRAS from EGFR or ALK, which are helpful to guide treatment and facilitate the discovery of new approaches capable of achieving this long-sought goal of cure in lung cancer patients with KRAS.

Sparse Deep Neural Network for Encoding and Decoding the Structural Connectome.

Brain state classification by applying deep learning techniques on neuroimaging data has become a recent topic of research. However, unlike domains where the data is low dimensional or there are large number of available training samples, neuroimaging data is high dimensional and has few training samples. To tackle these issues, we present a sparse feedforward deep neural architecture for encoding and decoding the structural connectome of the human brain. We use a sparsely connected element-wise multiplication as the first hidden layer and a fixed transform layer as the output layer. The number of trainable parameters and the training time is significantly reduced compared to feedforward networks. We demonstrate superior performance of this architecture in encoding the structural connectome implicated in Alzheimer's disease (AD) and Parkinson's disease (PD) from DTI brain scans. For decoding, we propose recursive feature elimination (RFE) algorithm based on DeepLIFT, layer-wise relevance propagation (LRP), and Integrated Gradients (IG) algorithms to remove irrelevant features and thereby identify key biomarkers associated with AD and PD. We show that the proposed architecture reduces 45.1% and 47.1% of the trainable parameters compared to a feedforward DNN with an increase in accuracy by 2.6 % and 3.1% for cognitively normal (CN) vs AD and CN vs PD classification, respectively. We also show that the proposed RFE method leads to a further increase in accuracy by 2.1% and 4% for CN vs AD and CN vs PD classification, while removing approximately 90% to 95% irrelevant features. Furthermore, we argue that the biomarkers (i.e., key brain regions and connections) identified are consistent with previous literature. We show that relevancy score-based methods can yield high discriminative power and are suitable for brain decoding. We also show that the proposed approach led to a reduction in the number of trainable network parameters, an increase in classification accuracy, and a detection of brain connections and regions that were consistent with earlier studies.