Automated obstructive sleep apnoea detection using symmetrically weighted local binary patterns
This Letter presents a computer-aided methodology for automated obstructive sleep apnoea (OSA) detection using the proposed symmetrically weighted local binary pattern (SLBP)-based features. The SLBP, which is a variant of one-dimensional local binary pattern (LBP), generates a binary pattern by making comparisons in the left and right neighbourhood of a sample. However, as opposed to LBP, the generated binary information is encoded into decimal value by using a symmetric weighting scheme. The proposed encoding scheme helps to reduce the length of the feature vector significantly. Experimental evaluations on the Physionet sleep apnoea single-lead electrocardiography signals suggest that the proposed SLBP features are effective in detecting OSA with an accuracy of 89.80%. Our results also show that the proposed SLBP achieves a good trade-off between the classification and computational performance among different variants of LBP. Further, the proposed approach outperforms recently proposed methodologies for OSA detection.