DEEP LEARNING METHOD FOR ENHANCING SLEEP STAGING CLASSIFICATION

Zhu, Zhengzhong

Permalink

https://hdl.handle.net/2142/124865 Copy

Description

Title

DEEP LEARNING METHOD FOR ENHANCING SLEEP STAGING CLASSIFICATION

Author(s)

Zhu, Zhengzhong

Issue Date

2022-05-01

Keyword(s)

Deep Learning; Sleep Staging Classification

Abstract

Traditionally, doctors classify the sleep staging of patients by manually examining the spectrogram of electroencephalogram (EEG) data at different time intervals. We propose a twostage pipeline that automatically predicts the staging of sleep given enough confidence and requests for manual labeling from doctors given enough uncertainty. In the first stage, multiple deep learning models, such as VGG-16 and ResNet-50, take in EEG data preprocessed with fast Fourier transform (FFT) and output a prediction and some confidence scores. In the second stage, multiple models with trainable parameters take in the confidence scores and output the decision of whether accept or reject the prediction. Among different combinations of models, RESNET-50 and SCRIB achieve the lowest average class-specific risk.

Type of Resource

text

Language

eng

Owning Collections