Abstract-Machine learning techniques are widely used now for neuro-imaging based diagnosis. These methods yield fully automated clinical decisions, unbiased by variable radiological expertise. This research paper compares and evaluates the performance and reliability of conventional Least Square Support Vector Machine (LSSVM) with that of Particle Swarm Optimization (PSO) based LSSVM in the diagnosis of dementia. The manual interpretation of large volume of brain MRI and cognitive measures may lead to incomplete diagnosis. The PSO-LSSVM approach is trained with multiple biomarkers to facilitate effective, accurate classification which is a requirement of the hour. Wavelet based texture features and multiple biomarkers are fed as input to the classifier. PSO-LSSVM yields 98% accurate results and outperforms LSSVM classifier in terms of sensitivity, specificity and accuracy in this analysisIndex Terms-Classification, dementia, least square support vector machine, particle swarm optimization. I. INTRODUCTIONMedical data and neuroimaging has increasingly employed techniques from machine learning and computer aided diagnostics. For instance, a set of training input data is made to yield a desired output by means of a supervised machine learning algorithm that is "trained" for the purpose. Automated classification methods are commonly used for the analysis of neuroimaging studies. Several multi-resolution approaches have been proposed to detect significant changes in the brain volume using neighbourhood information. Various computer-aided techniques have been proposed in the past and include the study of texture changes in signal intensity [1], grey matter (GM) concentration differences [2], atrophy of sub-cortical limbic structures [3]- [5], and general cortical atrophy [6]- [8].Brain image analyses have widely relied on univariate voxel-wise analyses, such as voxel-based morphometry (VBM) for structural MRI [9]. In such analyses, brain images are first spatially registered to a common stereotaxic space, and then mass univariate statistical tests are performed in each voxel to detect significant group differences. However, the sensitivity of these approaches is limited when the differences are spatially complex and involve a combination of different voxels or brain structures [10]. Recently, there has been a growing interest in support vector machines (SVM) Manuscript received December 30, 2012; revised April 11, 2013. T. R. Sivapriya is with the Department of Computer Science, Lady Doak College, Madurai, Tamilnadu, India. (e-mail:spriya.tr@gmail.com Research papers with different approaches for image classification and segmentation are reported in the literature. This study sets out to investigate the reliability and efficiency of PSO based LSSVM techniques in detecting demented and non-demented patients combining MMSE and Clinical Dementia Ratio (CDR) with brain volume. To achieve a high degree of success in treatment, one requirement is to establish proper classification and diagnosis of patients with complex neur...