A surface-based approach for classification of 3D neuroanatomic structures

Shape modeling is an important area in computer vision and medical imaging. Shape models have proved valuable in the tasks of object representation, recognition, and classification. This thesis focuses on shape models of 3D surfaces extracted from volumetric images and their applications in medical image analysis. This thesis presents a new framework for classification of 3D neuroanatomic structures using shape features, with the goal of deriving a shape-based medical classifier that has clinical value. This framework combines a powerful 3D surface modeling method, the spherical harmonics (SPHARM) description, with a set of effective pattern classification, feature selection, evaluation, and visualization techniques. This approach is shown to be effective using simulated shape data. It is also applied to real hippocampal data, and achieves good cross-validation accuracy (>90%) in detecting hippocampal shape changes in schizophrenia, which is competitive with the best results in previous studies. To help medical diagnosis in practice, a threshold-free receiver operating characteristic approach is employed as a means of evaluating the potential of a classifier. In addition, a new visualization approach is proposed for localizing discriminative patterns. This thesis also considers two extensions of the above work. In one extension, the left and right hippocampi are treated as a single shape configuration, and a new multi-surface alignment algorithm is developed for aligning configurations of multiple surfaces simultaneously. This alignment algorithm is incorporated into our classification for hippocampal data. The best cross-validation accuracy achieved is 92%, supporting the idea that the shape configuration of the two hippocampi provides valuable information in detecting schizophrenia. In the other extension, we focus on spherical parameterization, which is the key step for deriving a SPHARM description for a 3D object. Earlier techniques can be applied only to voxel surfaces. We propose CALD, a new spherical parameterization algorithm which makes our framework applicable to general triangle meshes. This new algorithm can be used for shape representation and manipulation purposes in different areas, including medical image analysis, computer vision, graphics, and multimedia databases.

A healthy liver leads to healthy life. In India, as well as in other parts of the world, liver disease is one of the principle areas of concern in medicine. For this study, diagnosis of liver disease is performed by deploying classification methods include linear discriminant analysis (LDA), quadratic discriminant analysis (QDA), feed-forward neural network (FFNN) and support vector machine (SVM) based approaches. Experimental results concluded that SVM based approaches outperformed all other classification methods in terms of diagnostic accuracy rates. Furthermore, least squares support vector machine (LSSVM) with gaussian radial basis kernel function based machine learning approach had emerged as the as the best predictive model by reducing inefficiencies caused by false diagnosis. LSSVM also performed better than linear SVM, polynomial SVM, quadratic SVM and multilayer perceptron SVM despite the uneven variance in attribute values in the health examination data.

A hand profile characterisation approach for biometric identification with contactless hand image acquisition is evaluated. The approach models the shapes of fingers with Point Distribution Models (PDMs), which consist of a mean shape and a number of eigenvectors which describe the main modes of variation of the shape class. The weighted PDM eigenvectors that capture the variation between the input finger shapes and the averaged finger shapes are used as feature vectors. Classification is performed using a least squares support vector machine. Experiments using multiple hand databases demonstrated the advantage of using finger PDMs.

Receiver operating characteristics (ROC) has been widely used as a performance evaluation tool to measure effectiveness of medical modalities. It is derived from a standard detection theory with false alarm and detection power interpreted as false positive (FP) and true positive (TP) respectively in terms of medical diagnosis. The ROC curve is plotted based on TP versus FP via hard decisions. This paper presents a three dimensional (3D) ROC analysis which extends the traditional two-dimensional (2D) ROC analysis by including a threshold parameter in a third dimension resulting from soft decisions, (SD). As a result, a 3D ROC curve can be plotted based on three parameters, TP, FP and SD. By virtue of such a 3D ROC curve three two-dimensional (2D) ROC curves can be derived, one of which is the traditional 2D ROC curve of TP versus FP with SD reduced to hard decision. In order to illustrate its utility in medical diagnosis, its application to magnetic resonance (MR) image classification is demonstrated.

The trend of current brain computer interfaces (BCI) seek to establish bi-directional communication with the brain, for instance, recovering motor functions by externally controlling devices and directly stimulating the brain. This will greatly assist paralyzed individuals through bypassing the damaged brain region. The key process of this communication interface is to decode movements from neural signals and encode information into neural activity. The majority of decoding or pattern classification studies have focused on cortical areas for BCIs, but deep brain structures have also been involved in motor control. The subthalamic nucleus (STN) in the basal ganglia is involved in the preparation, execution and imagining of movements, and may be an alternative source for driving BCIs. This study therefore aimed to classify patterns of deep brain local field potentials (LFPs) related to execution of visually cued movements. LFPs were recorded bilaterally from the STN through deep brain stimulation electrodes implanted in patients with Parkinson's disease. The frequency dependent components of the LFPs were extracted using the wavelet packet transform. In each frequency component, signal features were extracted using an alternative approach called neural synchronization by analyzing Granger causality between the STN. Based on these extracted features, a new feature selection strategy, namely weighted sequential feature selection (WSFS) was developed to efficiently select the optimal feature subset. A support vector machine (SVM) classifier was implemented alongside this novel feature extraction and selection strategy, and evaluated using a cross-validation procedure. Using this optimised feature subset, average correct pattern classification accuracy of movement (left or right) reached 76.0±3.1%. The results obtained in this study are encouraging and suggest that the neural activity in the deep neural circuit (basal ganglia) can be used for controlling BCIs.

Fluorescence molecular tomography in the near-infrared region is becoming a powerful modality for mapping the three-dimensional quantitative distributions of fluorochromes in live small animals. However, wider application of fluorescence molecular tomography still requires more accurate and stable reconstruction tools. We propose a shape-based reconstruction method that uses spherical harmonics parameterization, where fluorophores are assumed to be distributed as piecewise constants inside disjointed subdomains and the remaining background. The inverse problem is then formulated as a constrained nonlinear least-squares problem with respect to shape parameters, which decreases ill-posedness because of the significantly reduced number of unknowns. Since different shape parameters contribute differently to the boundary measurements, a two-step and modified block coordinate descent optimization algorithm is introduced to stabilize the reconstruction. We first evaluated our method using numerical simulations under various conditions for the noise level and fluorescent background; it showed significant superiority over conventional voxel-based methods in terms of the spatial resolution, reconstruction accuracy with regard to the morphology and intensity, and robustness against the initial estimated distribution. In our phantom experiment, our method again showed better spatial resolution and more accurate intensity reconstruction. Finally, the results of an in vivo experiment demonstrated its applicability to the imaging of mice.

The focus is made to develop predictor models for wind speed prediction employing the support vector machine neural models. Basically, support vector machines (SVM) is employed as classifiers, but this contribution models variant of SVM to act as predictors. A developed model of linear support vector machine (LSVM) and proximal support vector machine (PSVM) is proposed to carry out the wind speed prediction using the available real time wind farm data. In developed PSVM predictor, it is modeled for both linear PSVM predictor and non-linear PSVM predictor. The difference between the developed linear and non-linear PSVM predictor models lies in their applicability of kernel functions to perform effective wind speed prediction. The prediction application is implemented for the set of wind farm data with a wind mill height of 50 m in a manner to minimize the mean square error. The training process of the neural network algorithmic flow is done with the developed LSVM, L-PSVM (Linear PSVM) and N-PSVM (nonlinear PSVM) for predicting the wind speed in renewable energy systems. Results computed are compared with the other types of predictors to prove the effectiveness of the proposed variants of SVM predictors. The simulated results presents the effectiveness of the proposed predictors for the real time wind farm data and the applicability of the predictors for the considered datasets.

Boosting diagnosis accuracy of Alzheimer's disease using high dimensional recognition of longitudinal brain atrophy patterns

To develop and evaluate a model for obstructive sleep apnea (OSA) detection using an artificial neural network (ANN) based on the combined features of body mass index (BMI), electrocardiogram (ECG), and pulse oxygen saturation (SpO2). Polysomnography (PSG) data for 148 patients withOSA and 33 unaffected individuals were included. A multi-layer feed-forward neural network (FNN) was used based on the features obtained from ECG, SpO2, and BMI. The receiver operating characteristic (ROC) curve and the metrics of accuracy, sensitivity, and specificity were used to evaluate the performance of the overall classification. Some other machine learning methods including linear discriminant, linear Support Vector Machine (SVM), Complex Tree, RUSBoosted Trees, and Logistic Regression were also used to compare their performance with the FNN. The accuracy, sensitivity, and specificity of the proposed multi-layer FNN were 97.8%, 98.6%, and 93.9%, respectively, and the area under the ROC curve was 97.0%. Compared with the other machine learning methods mentioned above, the FNN achieved the highest performance. The satisfactory performance of the proposed FNN model for OSA detection indicated that it is reliable to screen potential patients withOSA using the combined channels of ECG and SpO2 and also taking into account BMI. This strategy might be a viable alternative method for OSA diagnosis.

Support Vector Machine (SVM) is a widely used supervised learning algorithm known for its robustness, efficiency, and theoretical reliability. This study analyzes the performance of linear kernel SVMs on real-world binary classification tasks from healthcare and social behavior domains. Two datasets were used: a heart disease dataset with 14 clinical features, and a social network advertisement dataset with demographic features like age and salary. Both were pre-processed using normalization, label encoding, and train-test splitting, and implemented in Python using scikit-learn. Evaluation metrics included accuracy, precision, recall, F1-score, and confusion matrix. The heart disease model achieved 83.3% accuracy, while the social network model reached 76.7%, showing linear SVM’s effectiveness on structured, moderately sized datasets. The study also covers core SVM concepts like support vectors, hyperplanes, and margin maximization. Soft margin classification was used to handle noisy data, and hyperparameter tuning (e.g., the regularization parameter C) improved performance. In conclusion, linear SVM is still a strong baseline for binary classification due to its simplicity, speed, and interpretability. Future work may explore multi-class problems, kernel comparisons, and ensemble methods to expand its applicability.

Fisher's linear discriminant (FLD), a feed-forward artificial neural network (ANN) and a support vector machine (SVM) were compared with respect to their ability to distinguish bursts from suppressions in electroencephalograms (EEG) displaying a burst–suppression pattern. Five features extracted from the EEG were used as inputs. The study was based on EEG signals from six full-term infants who had suffered from perinatal asphyxia, and the methods have been trained with reference data classified by an experienced electroencephalographer. The results are summarized as the area under the curve (AUC), derived from receiver operating characteristic (ROC) curves for the three methods. Based on this, the SVM performs slightly better than the others. Testing the three methods with combinations of increasing numbers of the five features shows that the SVM handles the increasing amount of information better than the other methods.

BackgroundPredictive models for delayed graft function (DGF) after kidney transplantation are usually developed using logistic regression. We want to evaluate the value of machine learning methods in the prediction of DGF.Methods497 kidney transplantations from deceased donors at the Ghent University Hospital between 2005 and 2011 are included. A feature elimination procedure is applied to determine the optimal number of features, resulting in 20 selected parameters (24 parameters after conversion to indicator parameters) out of 55 retrospectively collected parameters. Subsequently, 9 distinct types of predictive models are fitted using the reduced data set: logistic regression (LR), linear discriminant analysis (LDA), quadratic discriminant analysis (QDA), support vector machines (SVMs; using linear, radial basis function and polynomial kernels), decision tree (DT), random forest (RF), and stochastic gradient boosting (SGB). Performance of the models is assessed by computing sensitivity, positive predictive values and area under the receiver operating characteristic curve (AUROC) after 10-fold stratified cross-validation. AUROCs of the models are pairwise compared using Wilcoxon signed-rank test.ResultsThe observed incidence of DGF is 12.5 %. DT is not able to discriminate between recipients with and without DGF (AUROC of 52.5 %) and is inferior to the other methods. SGB, RF and polynomial SVM are mainly able to identify recipients without DGF (AUROC of 77.2, 73.9 and 79.8 %, respectively) and only outperform DT. LDA, QDA, radial SVM and LR also have the ability to identify recipients with DGF, resulting in higher discriminative capacity (AUROC of 82.2, 79.6, 83.3 and 81.7 %, respectively), which outperforms DT and RF. Linear SVM has the highest discriminative capacity (AUROC of 84.3 %), outperforming each method, except for radial SVM, polynomial SVM and LDA. However, it is the only method superior to LR.ConclusionsThe discriminative capacities of LDA, linear SVM, radial SVM and LR are the only ones above 80 %. None of the pairwise AUROC comparisons between these models is statistically significant, except linear SVM outperforming LR. Additionally, the sensitivity of linear SVM to identify recipients with DGF is amongst the three highest of all models. Due to both reasons, the authors believe that linear SVM is most appropriate to predict DGF.

In this study, we aimed to recognize six hand motions using surface electromyogram (sEMG) signals recorded from eight muscles of the right hand. Twenty-five healthy subjects participated in this experiment. Data were segmented using windows of 250-ms length with a 150-ms overlapping. In this paper, we extracted 24 features per muscle. Three feature sets-the original features, the features produced by a discriminant analysis (DA) and those selected by a multiple regression analysis (MRA) entered into one of the following classifiers: linear discriminant analysis (LDA) or support vector machine (SVM). The results showed that the original features classified by the SVM reached an average accuracy of 91.2 ± 0.383 %, significantly higher than the other approaches. The index finger extension (IFE) had higher classification accuracy than the other hand motions. The probability of the thumb opposition (TO) falsely classified as key pinch (KP) was 1.1 %, that of the hand grasp (HG) falsely classified as four fingers flexion (FFF) was 1.0 %.

We perceive the digital watermark detection as classification problem in image processing. We classify watermarked images as positive class whilst unwatermarked images as negative class. Support Vector Machine (SVM) is used as classifier of the watermarked and unwatermarked digital images. Two watermarking schemes i.e. Cox's spread spectrum (SS) and Single Value Decomposition (SVD) are used to embed watermark into digital images. These algorithms are selected based on their different level of robustness to Stirmark attacks. The payload of the watermark used for both algorithms is consistent at certain number of bits. SVM is trained with both the watermarked and unwatermarked images. Receiver Operating Characteristics (ROC) graphs are plotted to assess the statistical detection behavior of both the correlation detector and SVM classifier. We found that straight forward application of SVM leads to generalization problem. We suggest remedies to preprocess the training data in order to achieve substantially better performance from SVM classifier than those resulting from the straightforward application of SVM. Both watermarked and unwatermarked images are attacked under Stirmark and are then tested with the correlation detectors and SVM classifier. A comparison of the ROC of the correlation detectors and SVM classifier is performed to assess the accuracy of SVM classifier relative to correlation detectors. We found that SVM classifier has higher robustness to Stirmark attacks.

Objective: Support Vector Machine (SVM) is a machine-learning method, based on the principle of structural risk minimization, which performs well when applied to data outside the training set. In this paper, SVM was applied to predict 5-year survival status of patients with nasopharyngeal carcinoma (NPC) after treatment, we expect to find a new way for prognosis studies in cancer so as to assist right clinical decision for individual patient. Methods: Two modelling methods were used in the study; SVM network and a standard parametric logistic regression were used to model 5-year survival status. And the two methods were compared on a prospective set of patients not used in model construction via receiver operating characteristic (ROC) curve analysis. Results: The SVM1, trained with the 25 original input variables without screening, yielded a ROC area of 0.868, at sensitivity to mortality of 79.2% and the specificity of 94.5%. Similarly, the SVM2, trained with 9 input variables which were obtained by optimal input variable selection from the 25 original variables by logistic regression screening, yielded a ROC area of 0.874, at a sensitivity to mortality of 79.2% and the specificity of 95.6%, while the logistic regression yielded a ROC area of 0.751 at a sensitivity to mortality of 66.7% and gave a specificity of 83.5%. Conclusion: SVM found a strong pattern in the database predictive of 5-year survival status. The logistic regression produces somewhat similar, but better, results. These results show that the SVM models have the potential to predict individual patient's 5-year survival status after treatment, and to assist the clinicians for making a good clinical decision.