Paper Title
SACNN: Fast and Accurate Detection of Obstructive Sleep-Apnea using ECG Spectrograms

Abstract
Obstructive sleep apnea partially discontinues the passage of air for short time during sleep. In severe cases, it can lead to various cardiovascular and neurological complications such as hypertension and stroke. Research works available in the literature uncover the potential of machine learning and deep learning methods in the detection of sleep apnea using ECG signals. Although 2-D convolution neural network based models have proven better efficacy in the detection of sleep apnea, there is a huge scope to improve their detection accuracy without increasing the execution speed and increasing response time. In this study, the authors focus on deep learning based architectures and propose a ’Spatial Attention Convolution Neural Network’ for fast and accurate detection of sleep apnea. First, the Savitzky-Golay filter is used for noise suppression and Gabor Transformation is used for the conversion of 1-D ECG signal to a 2-D spectrogram. Thereafter, the proposed deep learning model uses the spectrogram for classification. The model utilizes spatial attention layers to refine the intermediate feature descriptors before passing them deeper into the network. Thus, simultaneously reducing the need of implementing a deep neural network. The Spatial Attention Convolution Neural Network reports the highest average accuracy of 96.97%, specificity of 96.75%, and sensitivity of 97.18% for sleep apnea detection. The results obtained prove that the proposed model outperforms other state-of-the-art 1-D and 2-D convolution neural network architectures. column. Keywords - Sleep Apnea, Spatial Attention, Signal Processing, Convolution Neural Networks, Classification