Extreme Learning Machine-Based Classification of ADHD Using Brain Structural MRI Data

Background

Effective and accurate diagnosis of attention-deficit/hyperactivity disorder (ADHD) is currently of significant interest. ADHD has been associated with multiple cortical features from structural MRI data. However, most existing learning algorithms for ADHD identification contain obvious defects, such as time-consuming training, parameters selection, etc. The aims of this study were as follows: (1) Propose an ADHD classification model using the extreme learning machine (ELM) algorithm for automatic, efficient and objective clinical ADHD diagnosis. (2) Assess the computational efficiency and the effect of sample size on both ELM and support vector machine (SVM) methods and analyze which brain segments are involved in ADHD.

Methods

High-resolution three-dimensional MR images were acquired from 55 ADHD subjects and 55 healthy controls. Multiple brain measures (cortical thickness, etc.) were calculated using a fully automated procedure in the FreeSurfer software package. In total, 340 cortical features were automatically extracted from 68 brain segments with 5 basic cortical features. F-score and SFS methods were adopted to select the optimal features for ADHD classification. Both ELM and SVM were evaluated for classification accuracy using leave-one-out cross-validation.

Results

We achieved ADHD prediction accuracies of 90.18% for ELM using eleven combined features, 84.73% for SVM-Linear and 86.55% for SVM-RBF. Our results show that ELM has better computational efficiency and is more robust as sample size changes than is SVM for ADHD classification. The most pronounced differences between ADHD and healthy subjects were observed in the frontal lobe, temporal lobe, occipital lobe and insular.

Conclusion

Our ELM-based algorithm for ADHD diagnosis performs considerably better than the traditional SVM algorithm. This result suggests that ELM may be used for the clinical diagnosis of ADHD and the investigation of different brain diseases.     

Rough-Fuzzy Clustering and Unsupervised Feature Selection for Wavelet Based MR Image Segmentation

Image segmentation is an indispensable process in the visualization of human tissues, particularly during clinical analysis of brain magnetic resonance (MR) images. For many human experts, manual segmentation is a difficult and time consuming task, which makes an automated brain MR image segmentation method desirable. In this regard, this paper presents a new segmentation method for brain MR images, integrating judiciously the merits of rough-fuzzy computing and multiresolution image analysis technique. The proposed method assumes that the major brain tissues, namely, gray matter, white matter, and cerebrospinal fluid from the MR images are considered to have different textural properties. The dyadic wavelet analysis is used to extract the scale-space feature vector for each pixel, while the rough-fuzzy clustering is used to address the uncertainty problem of brain MR image segmentation. An unsupervised feature selection method is introduced, based on maximum relevance-maximum significance criterion, to select relevant and significant textural features for segmentation problem, while the mathematical morphology based skull stripping preprocessing step is proposed to remove the non-cerebral tissues like skull. The performance of the proposed method, along with a comparison with related approaches, is demonstrated on a set of synthetic and real brain MR images using standard validity indices.     

Toward Probabilistic Diagnosis and Understanding of Depression Based on Functional MRI Data Analysis with Logistic Group LASSO

Diagnosis of psychiatric disorders based on brain imaging data is highly desirable in clinical applications. However, a common problem in applying machine learning algorithms is that the number of imaging data dimensions often greatly exceeds the number of available training samples. Furthermore, interpretability of the learned classifier with respect to brain function and anatomy is an important, but non-trivial issue. We propose the use of logistic regression with a least absolute shrinkage and selection operator (LASSO) to capture the most critical input features. In particular, we consider application of group LASSO to select brain areas relevant to diagnosis. An additional advantage of LASSO is its probabilistic output, which allows evaluation of diagnosis certainty. To verify our approach, we obtained semantic and phonological verbal fluency fMRI data from 31 depression patients and 31 control subjects, and compared the performances of group LASSO (gLASSO), and sparse group LASSO (sgLASSO) to those of standard LASSO (sLASSO), Support Vector Machine (SVM), and Random Forest. Over 90% classification accuracy was achieved with gLASSO, sgLASSO, as well as SVM; however, in contrast to SVM, LASSO approaches allow for identification of the most discriminative weights and estimation of prediction reliability. Semantic task data revealed contributions to the classification from left precuneus, left precentral gyrus, left inferior frontal cortex (pars triangularis), and left cerebellum (c rus1). Weights for the phonological task indicated contributions from left inferior frontal operculum, left post central gyrus, left insula, left middle frontal cortex, bilateral middle temporal cortices, bilateral precuneus, left inferior frontal cortex (pars triangularis), and left precentral gyrus. The distribution of normalized odds ratios further showed, that predictions with absolute odds ratios higher than 0.2 could be regarded as certain.     

Multi-method analysis of MRI images in early diagnostics of Alzheimer's disease

The role of structural brain magnetic resonance imaging (MRI) is becoming more and more emphasized in the early diagnostics of Alzheimer's disease (AD). This study aimed to assess the improvement in classification accuracy that can be achieved by combining features from different structural MRI analysis techniques. Automatically estimated MR features used are hippocampal volume, tensor-based morphometry, cortical thickness and a novel technique based on manifold learning. Baseline MRIs acquired from all 834 subjects (231 healthy controls (HC), 238 stable mild cognitive impairment (S-MCI), 167 MCI to AD progressors (P-MCI), 198 AD) from the Alzheimer's Disease Neuroimaging Initiative (ADNI) database were used for evaluation. We compared the classification accuracy achieved with linear discriminant analysis (LDA) and support vector machines (SVM). The best results achieved with individual features are 90% sensitivity and 84% specificity (HC/AD classification), 64%/66% (S-MCI/P-MCI) and 82%/76% (HC/P-MCI) with the LDA classifier. The combination of all features improved these results to 93% sensitivity and 85% specificity (HC/AD), 67%/69% (S-MCI/P-MCI) and 86%/82% (HC/P-MCI). Compared with previously published results in the ADNI database using individual MR-based features, the presented results show that a comprehensive analysis of MRI images combining multiple features improves classification accuracy and predictive power in detecting early AD. The most stable and reliable classification was achieved when combining all available features.     

Combined Metaheuristic Algorithm and Radiomics Strategy for the Analysis of Neuroanatomical Structures in Schizophrenia and Schizoaffective Disorders

ObjectivesSchizophrenia (SZ) is the most chronic disabling psychotic brain disorder. It is characterized by delusions and auditory hallucinations, as well as impairments in memory. Schizoaffective (SA) signs are co-morbid with SZ and are characterized by symptoms of SZ and mood disorder. Various researches suggest that SZ and SA share a number of equally severe cognitive deficits, but the pathophysiology has not yet been addressed in a comprehensive way. In this work, the heterogeneity in whole brain, ventricle and cerebellum region from psychotic MR brain images is examined using Machine learning and radiomic features.Materials and methodsT1 weighted MR brain images are obtained from Schizconnect database for the analysis. The shape prior level set method is used to segment the ventricle and cerebellum structures. The radiomic features which include shape and texture are extracted from these regions to discriminate the SZ and SA subjects. The performance of these features is evaluated with Binary Particle Swarm Optimization (BPSO) based Fuzzy Support Vector Machine (FSVM) classifier.ResultsThe shape constrained Level Set method is able to better segment ventricles and cerebellum regions from the images. The significant features that are extracted from whole brain, ventricle and cerebellum are identified by the BPSO based FSVM. The combination of radiomic features extracted from cerebellum region achieved high classification accuracy (90.09%) using metaheuristic algorithm. The extracted features from cerebellum are correlated with PANSS score. The causal analysis shows that there is an association been the tissue texture variation in identifying the disease severity. The symmetry analysis shows that left brain mean area is larger than the right side area. In particular SA has low cerebellum area compared to SZ. The radiomic features such as Hermite, Laws and tensor extracted from the left cerebellum show a significant texture variation in all the considered subjects (p<0.0001).ConclusionsThe results are clinically relevant in discriminating the pattern change in the structure, hence this biomarker and frame work could be used for the severity study of psychotic disorders.     

Evaluation of radiomic texture feature error due to MRI acquisition and reconstruction: A simulation study utilizing ground truth

The purpose of this study was to examine the dependence of image texture features on MR acquisition parameters and reconstruction using a digital MR imaging phantom. MR signal was simulated in a parallel imaging radiofrequency coil setting as well as a single element volume coil setting, with varying levels of acquisition noise, three acceleration factors, and four image reconstruction algorithms. Twenty-six texture features were measured on the simulated images, ground truth images, and clinical brain images. Subtle algorithm-dependent errors were observed on reconstructed phantom images, even in the absence of added noise. Sources of image error include Gibbs ringing at image edge gradients (tissue interfaces) and well-known artifacts due to high acceleration; two of the iterative reconstruction algorithms studied were able to mitigate these image errors. The difference of the texture features from ground truth, and their variance over reconstruction algorithm and parallel imaging acceleration factor, were compared to the clinical “effect size”, i.e., the feature difference between high- and low-grade tumors on T1- and T2-weighted brain MR images of twenty glioma patients. The measured feature error (difference from ground truth) was small for some features, but substantial for others. The feature variance due to reconstruction algorithm and acceleration factor were generally smaller than the clinical effect size. Certain texture features may be preserved by MR imaging, but adequate precautions need to be taken regarding their validity and reliability. We present a general simulation framework for assessing the robustness and accuracy of radiomic textural features under various MR acquisition/reconstruction scenarios.     

Colorization and automated segmentation of human T2 MR brain images for characterization of soft tissues

Characterization of tissues like brain by using magnetic resonance (MR) images and colorization of the gray scale image has been reported in the literature, along with the advantages and drawbacks. Here, we present two independent methods; (i) a novel colorization method to underscore the variability in brain MR images, indicative of the underlying physical density of bio tissue, (ii) a segmentation method (both hard and soft segmentation) to characterize gray brain MR images. The segmented images are then transformed into color using the above-mentioned colorization method, yielding promising results for manual tracing. Our color transformation incorporates the voxel classification by matching the luminance of voxels of the source MR image and provided color image by measuring the distance between them. The segmentation method is based on single-phase clustering for 2D and 3D image segmentation with a new auto centroid selection method, which divides the image into three distinct regions (gray matter (GM), white matter (WM), and cerebrospinal fluid (CSF) using prior anatomical knowledge). Results have been successfully validated on human T2-weighted (T2) brain MR images. The proposed method can be potentially applied to gray-scale images from other imaging modalities, in bringing out additional diagnostic tissue information contained in the colorized image processing approach as described.     

Impact of quantization algorithm and number of gray level intensities on variability and repeatability of low field strength magnetic resonance image-based radiomics texture features

PurposeThe purpose of this work was to investigate the impact of quantization preprocessing parameter selection on variability and repeatability of texture features derived from low field strength magnetic resonance (MR) images.MethodsTexture features were extracted from low field strength images of a daily image QA phantom with four texture inserts. Feature variability over time was quantified using all combinations of three quantization algorithms and four different numbers of gray level intensities. In addition, texture features were extracted using the same combinations from the low field strength MR images of the gross tumor volume (GTV) and left kidney of patients with repeated set up scans. The impact of region of interest (ROI) preprocessing on repeatability was investigated with a test-retest study design.ResultsThe phantom ROIs quantized to 64 Gy level intensities using the histogram equalization method resulted in the greatest number of features with the least variability. There was no clear method that resulted in the highest repeatability in the GTV or left kidney. However, eight texture features extracted from the GTV were repeatable regardless of ROI processing combination.ConclusionLow field strength MR images can provide a stable basis for texture analysis with ROIs quantized to 64 Gy levels using histogram equalization, but there is no clear optimal combination for repeatability.     

Searching for biomarkers of heart failure in the mass spectra of blood plasma

We have developed a technique for analysing blood plasma using MALDI-MS with subsequent data analysis to identify significant and specific differences between heart failure (HF) patients and healthy individuals. A training dataset comprising 100 HF patients and 100 healthy individuals was used to search for biomarkers (m/z range 1000-10,000). EWP cartridges when used in tandem with microcon centrifugal filters were found to give the best results. A data management chain including event binning, background subtraction and feature extraction was developed to reduce the data, and statistical analysis was used to map feature intensities on to a common scale. Various mathematical approaches including a simple cumulative score, support vector machines (SVM) and genetic algorithms (GAs) were then used to combine the results from individual features and provide a robust classification algorithm. The SVM gave the most promising results (accuracy 95%, receiver operating characteristic (ROC) score of 0.997 using 18 selected features). Finally, a test dataset comprising a further 32 HF patients and 20 controls was used to verify that the 18 putative biomarkers and classification algorithms gave reliable predictions (accuracy 88.5%, ROC score 0.998).     

Monitoring <i>Thosea sinensis</i> Walker in Tea Plantations Based on UAV Multi-Spectral Image

Thosea sinensis Walker (TSW) rapidly spreads and severely damages the tea plants. Therefore, finding a reliable operational method for identifying the TSW-damaged areas via remote sensing has been a focus of a research community. Such methods also enable us to calculate the precise application of pesticides and prevent the subsequent spread of the pests. In this work, based on the unmanned aerial vehicle (UAV) platform, five band images of multispectral red-edge camera were obtained and used for monitoring the TSW in tea plantations. By combining the minimum redundancy maximum relevance (mRMR) with the selected spectral features, a comprehensive spectral selection strategy was proposed. Then, based on the selected spectral features, three classic machine learning algorithms, including random forest (RF), support vector machine (SVM), and k-nearest neighbors (KNN) were used to construct the pest monitoring model and were evaluated and compared. The results showed that the strategy proposed in this work obtained ideal monitoring accuracy by only using the combination of a few optimized features (2 or 4). In order to differentiate the healthy and TSW-damaged areas (2-class model), the monitoring accuracies of all the three models were computed, which were above 96%. The RF model used the least number of features, including only SAVI and Band_red. In order to further discriminate the pest incidence levels (3-class model), the monitoring accuracies of all the three models were computed, which were above 80%, among which the RF algorithm based on SAVI, Band_red, VARI_green, and Band_{red_edge} features achieve the highest accuracy (OAA of 87%, and Kappa of 0.79). Considering the computational cost and model accuracy, this work recommends the RF model based on a few optimal feature combinations to monitor and distinguish the severity of TSW in tea plantations. According to the UAV remote sensing mapping results, the TSW infestation exhibited an aggregated distribution pattern. The spatial information of occurrence and severity can offer effective guidance for precise control of the pest. In addition, the relevant methods provide a reference for monitoring other leaf-eating pests, effectively improving the management level of plant protection in tea plantations, and guaranting the yield and quality of tea plantations.     

DROP: an SVM domain linker predictor trained with optimal features selected by random forest

MOTIVATION: Biologically important proteins are often large, multidomain proteins, which are difficult to characterize by high-throughput experimental methods. Efficient domain/boundary predictions are thus increasingly required in diverse area of proteomics research for computationally dissecting proteins into readily analyzable domains. RESULTS: We constructed a support vector machine (SVM)-based domain linker predictor, DROP (Domain linker pRediction using OPtimal features), which was trained with 25 optimal features. The optimal combination of features was identified from a set of 3000 features using a random forest algorithm complemented with a stepwise feature selection. DROP demonstrated a prediction sensitivity and precision of 41.3 and 49.4%, respectively. These values were over 19.9% higher than those of control SVM predictors trained with non-optimized features, strongly suggesting the efficiency of our feature selection method. In addition, the mean NDO-Score of DROP for predicting novel domains in seven CASP8 FM multidomain proteins was 0.760, which was higher than any of the 12 published CASP8 DP servers. Overall, these results indicate that the SVM prediction of domain linkers can be improved by identifying optimal features that best distinguish linker from non-linker regions.     

Accurate sequence-based prediction of catalytic residues

MOTIVATION: Prediction of catalytic residues provides useful information for the research on function of enzymes. Most of the existing prediction methods are based on structural information, which limits their use. We propose a sequence-based catalytic residue predictor that provides predictions with quality comparable to modern structure-based methods and that exceeds quality of state-of-the-art sequence-based methods. RESULTS: Our method (CRpred) uses sequence-based features and the sequence-derived PSI-BLAST profile. We used feature selection to reduce the dimensionality of the input (and explain the input) to support vector machine (SVM) classifier that provides predictions. Tests on eight datasets and side-by-side comparison with six modern structure- and sequence-based predictors show that CRpred provides predictions with quality comparable to current structure-based methods and better than sequence-based methods. The proposed method obtains 15-19% precision and 48-58% TP (true positive) rate, depending on the dataset used. CRpred also provides confidence values that allow selecting a subset of predictions with higher precision. The improved quality is due to newly designed features and careful parameterization of the SVM. The features incorporate amino acids characterized by the highest and the lowest propensities to constitute catalytic residues, Gly that provides flexibility for catalytic sites and sequence motifs characteristic to certain catalytic reactions. Our features indicate that catalytic residues are on average more conserved when compared with the general population of residues and that highly conserved amino acids characterized by high catalytic propensity are likely to form catalytic sites. We also show that local (with respect to the sequence) hydrophobicity contributes towards the prediction.     

Microarray data classification using automatic SVM kernel selection

Microarray data classification is one of the most important emerging clinical applications in the medical community. Machine learning algorithms are most frequently used to complete this task. We selected one of the state-of-the-art kernel-based algorithms, the support vector machine (SVM), to classify microarray data. As a large number of kernels are available, a significant research question is what is the best kernel for patient diagnosis based on microarray data classification using SVM? We first suggest three solutions based on data visualization and quantitative measures. Different types of microarray problems then test the proposed solutions. Finally, we found that the rule-based approach is most useful for automatic kernel selection for SVM to classify microarray data.     

Decoding Multiple Sound Categories in the Human Temporal Cortex Using High Resolution fMRI

Perception of sound categories is an important aspect of auditory perception. The extent to which the brain’s representation of sound categories is encoded in specialized subregions or distributed across the auditory cortex remains unclear. Recent studies using multivariate pattern analysis (MVPA) of brain activations have provided important insights into how the brain decodes perceptual information. In the large existing literature on brain decoding using MVPA methods, relatively few studies have been conducted on multi-class categorization in the auditory domain. Here, we investigated the representation and processing of auditory categories within the human temporal cortex using high resolution fMRI and MVPA methods. More importantly, we considered decoding multiple sound categories simultaneously through multi-class support vector machine-recursive feature elimination (MSVM-RFE) as our MVPA tool. Results show that for all classifications the model MSVM-RFE was able to learn the functional relation between the multiple sound categories and the corresponding evoked spatial patterns and classify the unlabeled sound-evoked patterns significantly above chance. This indicates the feasibility of decoding multiple sound categories not only within but across subjects. However, the across-subject variation affects classification performance more than the within-subject variation, as the across-subject analysis has significantly lower classification accuracies. Sound category-selective brain maps were identified based on multi-class classification and revealed distributed patterns of brain activity in the superior temporal gyrus and the middle temporal gyrus. This is in accordance with previous studies, indicating that information in the spatially distributed patterns may reflect a more abstract perceptual level of representation of sound categories. Further, we show that the across-subject classification performance can be significantly improved by averaging the fMRI images over items, because the irrelevant variations between different items of the same sound category are reduced and in turn the proportion of signals relevant to sound categorization increases.     

Analysis of recursive gene selection approaches from microarray data

MOTIVATION: Finding a small subset of most predictive genes from microarray for disease prediction is a challenging problem. Support vector machines (SVMs) have been found to be successful with a recursive procedure in selecting important genes for cancer prediction. However, it is not well understood how much of the success depends on the choice of the specific classifier and how much on the recursive procedure. We answer this question by examining multiple classifers [SVM, ridge regression (RR) and Rocchio] with feature selection in recursive and non-recursive settings on three DNA microarray datasets (ALL-AML Leukemia data, Breast Cancer data and GCM data). RESULTS: We found recursive RR most effective. On the AML-ALL dataset, it achieved zero error rate on the test set using only three genes (selected from over 7000), which is more encouraging than the best published result (zero error rate using 8 genes by recursive SVM). On the Breast Cancer dataset and the two largest categories of the GCM dataset, the results achieved by recursive RR are also very encouraging. A further analysis of the experimental results shows that different classifiers penalize redundant features to different extent and this property plays an important role in the recursive feature selection process. RR classifier tends to penalize redundant features to a much larger extent than the SVM does. This may be the reason why recursive RR has a better performance in selecting genes.     

Development of CYP3A4 inhibition models: comparisons of machine-learning techniques and molecular descriptors

Computational models of cytochrome P450 3A4 inhibition were developed based on high-throughput screening data for 4470 proprietary compounds. Multiple models differentiating inhibitors (IC(50) <3 microM) and noninhibitors were generated using various machine-learning algorithms (recursive partitioning [RP], Bayesian classifier, logistic regression, k-nearest-neighbor, and support vector machine [SVM]) with structural fingerprints and topological indices. Nineteen models were evaluated by internal 10-fold cross-validation and also by an independent test set. Three most predictive models, Barnard Chemical Information (BCI)-fingerprint/SVM, MDL-keyset/SVM, and topological indices/RP, correctly classified 249, 248, and 236 compounds of 291 noninhibitors and 135, 137, and 147 compounds of 179 inhibitors in the validation set. Their overall accuracies were 82%, 82%, and 81%, respectively. Investigating applicability of the BCI/SVM model found a strong correlation between the predictive performance and the structural similarity to the training set. Using Tanimoto similarity index as a confidence measurement for the predictions, the limitation of the extrapolation was 0.7 in the case of the BCI/SVM model. Taking consensus of the 3 best models yielded a further improvement in predictive capability, kappa = 0.65 and accuracy = 83%. The consensus model could also be tuned to minimize either false positives or false negatives depending on the emphasis of the screening.     

Identification of soybean based on Sentinel-1/2 SAR and MSI imagery under a complex planting structure

Soybeans, as the major global oil seed crops, are among the large-scale agricultural products imported into China. In this way, the accurate identification of their growing areas is the basis for agricultural decision-making and planting structure adjustment, which are of great significance to the national food security. Currently, there is insufficient research on the identification of soybean, using remote sensing, under complex agricultural planting structures. In this paper, Guoyang County, a typical soybean-producing region in the north of Anhui Province, China, was selected as the study area, and then the multi-temporal Sentinel-1/2 (S-1/2) microwave and optical multispectral data were integrated to obtain the spatial distribution of the planting areas, via a stepwise hierarchical extraction strategy. Considering the Jeffries-Matusita (JM) distance, August 18 (i.e., the early pod setting stage) was correspondingly determined as more appropriate for soybean extraction, compared with other time phases. For this purpose, a set of rules were established to eliminate the non-cropland pixels, and the total distribution of the field vegetation was derived accordingly. In total, 30 candidate features (viz. 10 spectral bands, 11 vegetation indices, 4 texture features, and 5 microwave polarization features) were selected, and their importance was evaluated based on the ReliefF algorithm. With the support of the typical cover type samples, the weight assessment method for the ReliefF feature was further bundled with three machine-learning (ML) algorithms, namely, the random forest (RF), the back-propagation neural network (BPNN), and the support vector machine (SVM), to single out the optimum subset of the features for soybean identification. The results finally revealed that the ReliefF-RF model outperformed the other two. With nine optimum features, Kappa coefficient also reached up to 0.77–0.85. Furthermore, the extraction accuracy produced by the optimum subset of the features was significantly greater than the original 10 spectral bands (Kappa coefficient of 0.75–0.84), albeit slightly lower than the effect of the total 30 features (Kappa coefficient of 0.78–0.85).     

Feature Selection and the Class Imbalance Problem in Predicting Protein Function from Sequence

When the standard approach to predict protein function by sequence homology fails, other alternative methods can be used that require only the amino acid sequence for predicting function. One such approach uses machine learning to predict protein function directly from amino acid sequence features. However, there are two issues to consider before successful functional prediction can take place: identifying discriminatory features, and overcoming the challenge of a large imbalance in the training data. We show that by applying feature subset selection followed by undersampling of the majority class, significantly better support vector machine (SVM) classifiers are generated compared with standard machine learning approaches. As well as revealing that the features selected could have the potential to advance our understanding of the relationship between sequence and function, we also show that undersampling to produce fully balanced data significantly improves performance. The best discriminating ability is achieved using SVMs together with feature selection and full undersampling; this approach strongly outperforms other competitive learning algorithms. We conclude that this combined approach can generate powerful machine learning classifiers for predicting protein function directly from sequence.