Brain Computer Interface (BCI) technology is recently being spotlighted as the performance of Artificial Intelligence (AI) increased drastically. Although there are several instruments to measure brain's activity, Electroencephalography(EEG) signal is in limelight as it has high temporal resolution and is non-invasive, while being extremely portable. However, the raw EEG signals can not be directly used in BCI systems as it includes a lot of artifacts, and has numerous channels. Therefore, selecting only the needed channels and rejecting ones that include big noises are crucial to increase the performance of classification. Motor Imagery EEG (MI-EEG) is a type of EEG where subjects imagine that they are moving their body. MI-EEG channel selection techniques can be divided into two big groups: Common Spatial Pattern (CSP) based models and non-CSP based models. Therefore in this paper, we introduce the models classified in the criteria above, using detailed and strict mathematical terms, and compare each approach's accuracies.