Automated Diagnosis of Neonatal Encephalopathy on aEEG Using Deep Neural Networks

Publication date: Available online 20 January 2020Source: NeurocomputingAuthor(s): Jingling Wang, Rong Ju, Yuanyuan Chen, Guijun Liu, Zhang YiAbstractAmplitude-integrated electroencephalography, called aEEG clinically, is an effective approach for monitoring neonatal cerebral function. Early aEEG screening is necessary to diagnose neonatal encephalopathy. The electroencephalography (EEG) signal contains lots of information about infant neurological status and prognosis. Traditional computer-aided diagnosis methods pay more attention to process EEG signal data, while doctors referring more to aEEG screening images in clinical diagnosis. In addition, clinical diagnosis of neonatal encephalopathy is a large challenge in low-level medical resource settings where few neonatologists are available to care for those newborns with cerebral disease. As a result, a method of automated diagnosing EEG-related images is requested. This paper presents an automated neonatal encephalopathy diagnosis model for aEEG images, which is divided into two sub-networks. The front part is utilized to extract high-dimensional features from multiple parts of an image. And the second portion aims to accomplish classification. There is a feature combination operator between the two parts. The feature enhancing mechanism can protrude the important region and achieve a better result. The annotated datasets, containing more than 600 aEEG screening images, are established to train and evaluate the novel model....
Source: Neurocomputing - Category: Neuroscience Source Type: research