ABSTRACT

Lateralization accuracy determines the success of resective surgery in medically refractory mesial temporal lobe epilepsy (mTLE). For presurgical evaluations, various imaging modalities, including magnetic resonance imaging (MRI), have been used. In recent years, quantitative analysis of MRI along with machine learning models have been proposed to enhance diagnosis and treatment of mTLE patients. An optimized machine learning approach for lateralizing mTLE patients consists of two major steps: feature engineering—feature extraction, selection, and reduction—and classification. In this chapter, various types of features extracted from MRI-based structural imaging modalities including T1-weighted images, fluid-attenuated inversion recovery images, and diffusion tensor imaging are described. The features are defined based on the image intensity and texture as well as morphology of specific brain structures related to mTLE. The rationale for the employed feature extraction, selection, and reduction approaches along with the classification methods are discussed. The results of employing each set of features and classifiers are presented. Distinctive contributions of each feature and classifier towards noninvasive mTLE lateralization are explained. Limitations of the existing methods and directions for future work are described.