Computer assisted interpretation of the human EEG: improving diagnostic efficiency and consistency in clinical reviews

Scalp electroencephalography (EEG) measures brain activity non-invasively by using electrodes on the scalp and capturing small electrical fluctuations caused by the firing of neurons. From these recordings, a clinical neurophysiologist can study the captured patterns and waveforms and determine if any abnormal brain activity exist. The disadvantage of EEG recordings are that they require an expert for interpretation and diagnosis. Not only this, but visual reviewing is also time consuming and a high degree of inter-rater variability exists between clinicians. Automated analysis by means of computerized algorithms can lessen the burden on visual reviews and provide more consistency in diagnostic reports. Given the complexity of the signal, automated analysis has only been partially implemented with limited clinical use. The main objective of this thesis was to find reliable computerized methods and efficient reviewing techniques that will assist with the review and interpretation of routine outpatient EEGs. To achieve this goal, algorithms were developed to automatically characterize five common EEG background properties: the posterior dominant peak frequency; reactivity; anterior-posterior gradients; symmetry; and the presence or absence of diffuse slow-wave activity. In addition to characterizing the background activity, an intuitive and efficient technique was also developed for the automated detection of inter-ictal epileptiform discharges. To evaluate our algorithms and reviewing techniques, a software application was developed and experienced neurologists and clinical neurophysiologists across the Netherlands were invited to evaluate and test the feasibility our approach. Very positive results were achieved. Additional testing and minor improvements are needed to bring this work io clinical practice, but the overall results show that the described methods and reviewing strategies, together with acceptance by clinicians, have been successfu