Pattern recognition in airflow recordings to assist in the sleep apnoea–hypopnoea syndrome diagnosis

Abstract

This paper aims at detecting sleep apnoea-hypopnoea syndrome (SAHS) from single-channel airflow (AF) recordings. The study involves 148 subjects. Our proposal is based on estimating the apnoea-hypopnoea index (AHI) after global analysis of AF, including the investigation of respiratory rate variability (RRV). We exhaustively characterize both AF and RRV by extracting spectral, nonlinear, and statistical features. Then, the fast correlation-based filter is used to select those relevant and non-redundant. Multiple linear regression, multi-layer perceptron (MLP), and radial basis functions are fed with the features to estimate AHI. A conventional approach, based on scoring apnoeas and hypopnoeas, is also assessed for comparison purposes. An MLP model trained with AF and RRV selected features achieved the highest agreement with the true AHI (intra-class correlation coefficient = 0.849). It also showed the highest diagnostic ability, reaching 92.5 % sensitivity, 89.5 % specificity and 91.5 % accuracy. This suggests that AF and RRV can complement each other to estimate AHI and help in SAHS diagnosis.

Assessment of Time and Frequency Domain Entropies to Detect Sleep Apnoea in Heart Rate Variability Recordings from Men and Women

“…It has been established that the typical duration of apnoeic events ranges from 20 to 40 s [45], which in the frequency domain would mainly affect the 0.025-0.050 Hz band. Latest studies confirmed this as the band with the highest statistically significant differences between SAHSpositive and SAHS-negative subjects in the airflow signal [46,47]. Additionally, a recent study has reported an increased cardio-respiratory coordination during the apnoeic events [48].…”

Pulse Rate Variability Analysis to Enhance Oximetry as at-Home Alternative for Sleep Apnea Diagnosing

“…MLP is a supervised learning algorithm typically arranged in three fully connected layers (input, hidden, and output) [13]. The layers are formed by neurons, each of them characterized by an activation function gð Þ and their connections (or weights) to neurons from other layers (w i;j , being i and j different layers).…”

“…Linear activation functions were also used for each neuron in the hidden layer ðN H Þ. The number of hidden neurons finally arranged is a tuning parameter optimized using the training set [13]. In order to prevent overfitting, a regularization parameter (a) was introduced during the MLP training process [13].…”

“…It was also optimized along with N H . All weights w i;j were optimized using the sum of squares error function minimization criterion by means of the scaled conjugate gradient algorithm [13].…”

“…The eight features extracted from NPO (SpO 2 and PRV) are subsequently used as the only source of information to train a multi-layer perceptron (MLP) artificial neural network with ability to automatically estimate AHI. We chose MLP in view of its success in previous studies focused on automatic AHI estimation [12,13]. The diagnostic performance of this estimated AHI is compared with ODI 3 to assess whether the use of PRV information can improve the ability of oximetry to detect SAHS and its severity.…”