The non-invasive Brain-Computer Interface (BCI) developed in our lab targets asynchronous operation of devices by monitoring electroencephalographic (EEG) activity and identifying oscillatory patterns that the user can voluntary modulate through the execution of motor imagery (MI) tasks. Successful self-paced interaction under this framework requires the incorporation of an evidence accumulation module to eliminate the uncertainty of single-sample classification and to drive an efficient feedback visualization. In this work, we motivate the need for this additional module, describe its role in a closed-loop MI BCI and present a comparative study of two different frameworks for evidence accumulation.