EEG Spectrograms to Determine Occurrence or Absence of Sleep Apnoea Using Deep Learning

Sleep apnoea has an increased mortality and morbidity. The diagnosis of the disease is cumbersome and very resource intensive. In a setup where the individual health is followed over a long period of time and records maintained, artificial intelligence-based model may help predict the presence of disease.

With only two inputs, oximetry and one channel EEG, a model was devised based on EEG spectrograms which were put through VGG 19 and custom-made convolutional neural network (CNN). This was done on one full night polysomnographic data available from open data source.

The model gave a prediction of almost 100%. An innovative method for screening of sleep apnoea is proposed using this model.

The pilot work of building the model, preprocessing of EEG data, conversion to EEG spectrogram, using VGG 19 model and custom-made CNN model is discussed. Deep Learning models might be useful in screening out such disorders.