Sleep-related breathing disorders are diseases related to pharyngeal airway collapse. It can lead to several health problems such as somnolence, poorer daytime cognitive performance, and cardiovascular morbidity and mortality. However, computer-aided diagnostic (CAD) tools play a very important role in the detection of breathing disorders. It is possible to measure breathing activity, but most approaches require some type of device placed on the human body. This paper proposes a novel methodology of an unobtrusive CAD system to the breathing disorder detection. Unobtrusive approach is ensured by ballistocardiography (BCG) sensors located on the measured bed. The significant pieces of information from the signals are extracted by Cartan curvatures. Thereafter, important features are separated from individual samples as an input to our 9-layer deep convolutional neural network. We achieved an average accuracy of 98.00%, sensitivity of 94.26%, and specificity of 99.22% on 4009 regular and 1307 disordered breathing samples.

睡眠相关的呼吸障碍是与咽气道塌陷有关的疾病。它可能导致一些健康问题，例如嗜睡，日间认知能力下降以及心血管疾病的发病率和死亡率。但是，计算机辅助诊断（CAD）工具在呼吸障碍的检测中起着非常重要的作用。可以测量呼吸活动，但是大多数方法都需要将某种类型的设备放在人体上。本文提出了一种用于呼吸障碍检测的不干扰性CAD系统的新方法。通过位于被测床上的心动描记法（BCG）传感器，可以确保不引人注目的方法。来自信号的重要信息是通过Cartan曲率提取的。之后，重要特征与单个样本分开，作为我们9层深度卷积神经网络的输入。我们对4009例常规和1307例呼吸紊乱样本的平均准确度为98.00％，灵敏度为94.26％，特异性为99.22％。