Detecting Epileptic Activities from resting tMRJ Time-course by using PCA algorithm Qiyi Song, Huafu Chen, Dezhong Yao School of Life Science & Technology University of Electronic Science and Technology of china Chengdu 610054,china E-mail:uestcsqyl 63.com chenhf@uestc.edu.cn Abstract-A delay principal component analysis (PCA) novel method is presented for detecting interictal epileptic activities from resting functional magnetic resonance (fMRI) time-course dataset. The proposed method consists of three steps: (1) calculating the covariance matrix with time delay and then conducting SVD on the matrix; (2) calculating the weighted correlation coefficient and applying T-test to every voxel; (3) only those voxel whose weighted correlation coefficient T-value are larger than the threshold (p<0.02) and whose mean and standard deviation belong to each special scope are considered as the epileptic foci. In contrast with traditional PCA methods, our algorithm further considers time delay and makes T-test for weighted correlation coefficient rather than voxel value. Computer simulation and vivo epilepsy fMRI data analysis demonstrate the potential of this technique to localize epilepsy foci. L. INTRODUCTION .Epilepsy is characterized by transient behavioral and electro-physiologic disturbance, which may or may not be associated with detectable structural lesions [1]. With the development of science, increasing epilepsy patients whose cases are refractory to medical intervention are benefited from surgical intervention. Successful outcome following resection for epilepsy depends on accurate localization of the epileptogenic zone. Since epilepsy is a functional disorder, functional tests, electroencephalography (EEG) in particular, have played a central role [2]. However, many of the functional localization methods currently used in the presurgical evaluation have limitations. In particular, EEG has relatively low spatial resolution, and both EEG and positron emission topography (PET) have limitations with respect to specificity of abnormalities [2]. There is clearly a need for development of a functional evaluation that can overcome these problems. Functional magnetic resonance imaging (fMRI) is one of the most significant and revolutionary advances in MRI in recent years. Noninvasive detection of blood oxygenation dependent (BOLD) contrast in human cortices by fMRI provides a powerful tool for cognitive and neurophysiological research [3]. As one of the most efficient Guangming Lu, Zhiqiang Zhang Department of Medical image, Nanjin general hospital of PLA, Nanjin, 210002,PR China E-mail:zhangzhiqiang@nju.org.cn methods with very high spatial resolution, fMRI is very important to precisely localize brain activities. In epilepsy, the aim is to localize the spontaneous brain electrical activities whose induction cannot be consistently controlled. This prompted the development of a technique known as EEG-triggered fMRI to localize epileptic brain activities seen on EEG [4]. In EEG-triggered fMRRL a specifically designed, MRI-compatible EEG system is used to monitor the epileptic patient for interracial spikes during the MRI scan. [5]. These procedures require MI and EEG system modifications to ensure patient safety and MR image quality [6], and they have significant shortcomings. Besides being cumbersome, they are useful only in patients with frequent interictal epileptiform discharges recorded from the scalp, and they are not sensitive to interictal epileptiform activities in deep structures, or interictal epileptiform discharges with a dipole orientation that is unfavorable for scalp recordings. It is well known that intracranial EEG records much epileptiform activities that are not noted on the scalp. Functional MRI is not limited by dipole orientation or spatial location; therefore it has the potential to overcome these limitations of EEG if it can be implemented independently to EEG. The challenge is to develop a method without EEG to accurately detect epileptiform discharges [2]. In the localization of epileptic foci by fMRI without EEG triggering, the timing of the stimuli (endogenous epileptic activities) throughout the imaging series is unclear, which is a resting fMRI problem, thus the model-driven methods such as the Statistical Parametric Mapping (SPM) can not be utilized. Recently, the temporal clustering analysis of fMRI data (TCA) [7] was developed for localizing interictal epileptic activity [9], but the interictal epileptic activities can be detected only in high intensity epileptic activities. The independent component analysis (ICA) and principal component analysis (PCA) are all data-driven methods without requirement of prior information about experiment, however, the ICA has an inevitable problem: how to select component from hundreds of components. Because it still hasn't a proper criterion for localizing epileptic foci, so far, it makes unreliable the subjective decisions. These defects 0-7803-9422-4/05/$20.00 C2005 IEEE 1552