0949 Real-time acoustic apnea event detector by training a deep learning model with home noise added data
نویسندگان
چکیده
Abstract Introduction For diagnosis and management of Obstructive Sleep Apnea (OSA), long-term multi-night monitoring is crucial. Convenient detection OSA at home required for this purpose. Using sound recorded by smartphone can provide a convenient way to detect OSA. In study, we present sound-based deep learning model that in real-time even environment where various noises exist. The trained with noise simulated be robust detecting noises. Methods Two types data were used training testing. first type was sleep breathing collected the hospital while patients underwent PSG. It included 1,154 297 nights PSG microphone smartphone, respectively. We split them into 150 testing rest training. second data, which 22,500 might occur residential environment. proposed acoustic apnea event detector inputs Mel spectrograms sounds outputs classes each epoch (APNEA, HYPOPNEA, or NO-EVENT). make noisy performance prediction assessed epoch-by-epoch accuracy severity classification based on apnea-hypopnea index (AHI). Results Our achieved 86 % agreement (0.75 macro F1) 3-class task. had an 92% NO-EVENT, 84% APNEA, 51% HYPOPNEA. Most misclassifications made 15% 34% HYPOPNEA being wrongly predicted as APNEA sensitivity specificity (AHI ≥ 15) 0.85 0.84, Conclusion study presents works variety environments. Based this, additional research needed verify usefulness diagnostic technologies Support (if any)
منابع مشابه
Melanoma detection with a deep learning model
Background: Skin cancer is one of the most common forms of cancer in the world and melanoma is the deadliest type of skin cancer. Both melanoma and melanocytic nevi begin in melanocytes (cells that produce melanin). However, melanocytic nevi are benign whereas melanoma is malignant. This work proposes a deep learning model for classification of these two lesions. Methods: In this analytic s...
متن کاملSemi-Supervised Training in Deep Learning Acoustic Model
We studied the semi-supervised training in a fully connected deep neural network (DNN), unfolded recurrent neural network (RNN), and long short-term memory recurrent neural network (LSTM-RNN) with respect to the transcription quality, the importance data sampling, and the training data amount. We found that DNN, unfolded RNN, and LSTM-RNN are increasingly more sensitive to labeling errors. For ...
متن کاملSimulation of a Neutron Detector for Real Time Imaging Applications
Monte Carlo Method is used to simulate a double layer gadolinium-amorphous silicon thermal neutron detector. The detector fabricated in pixel array configuration has various applications including neutron imaging. According to the simulation results, a detector consisting of a gadolinium (Gd) film with thickness of 2-4 ~m, sandwiched properly with two layers of sufficiently thick (-30 ?µm) hydr...
متن کاملa study on insurer solvency by panel data model: the case of iranian insurance market
the aim of this thesis is an approach for assessing insurer’s solvency for iranian insurance companies. we use of economic data with both time series and cross-sectional variation, thus by using the panel data model will survey the insurer solvency.
ذخیره در منابع من
با ذخیره ی این منبع در منابع من، دسترسی به آن را برای استفاده های بعدی آسان تر کنید
ژورنال
عنوان ژورنال: Sleep
سال: 2023
ISSN: ['0302-5128']
DOI: https://doi.org/10.1093/sleep/zsad077.0949