Application of hidden Markov algorithm to the localization of epilepsy focus

Epilepsy is a chronic neurological disorder characterized by recurrent seizures that often occur suddenly and without warning. This condition affects individuals of all ages and is among the most prevalent neurological disorders, making its diagnosis and treatment a significant focus of medical research. In this study, we employed a Hidden Markov Model (HMM) to localize epileptic foci. First, a brain model was constructed using magnetic resonance imaging (MRI) data. Subsequently, statistical models of brain activity in various states were developed for patients using high-density scalp electroencephalogram (HD-EEG) data and the HMM, which allowed us to obtain state time probability sequences. Linear constrained minimum variance (LCMV) beamforming was then applied, in combination with the state time probability sequences, to identify the spatial characteristics of different brain states. Virtual electrodes were constructed to determine epileptiform states based on the peak values of these virtual electrodes. Finally, the identified epileptogenic areas were visualized on MRI images and compared with clinical stereoelectroencephalography (SEEG) electrode coordinates to assess localization distance. The positioning distance for this method in 23 patients was 13.93 mm ± 5.78 mm, smaller than the 16.18 mm ± 3.78 mm distance range of the Equivalent Current Dipole (ECD) model in sLORETA algorithm. For the 23 patients with multiple lesions, this approach also enabled us to determine the correlation between the discharge activities at each focal point using the state transition probability matrix, thereby identifying the origin and propagation nodes of abnormal epileptic discharges. This method allows for detailed visualization and analysis, enabling physicians to more accurately identify and localize focal areas, leading to more effective treatment strategies.