Low-complexity image processing for real-time detection of neonatal clonic seizures

Area of Science:

• Medical Imaging
• Neonatal Neurology
• Biomedical Signal Processing

Background:

• Neonatal clonic seizures are a significant concern in newborns, requiring accurate and timely detection.
• Existing detection methods may have limitations in real-time application or complexity.
• Image processing offers a non-invasive modality for monitoring neonatal physiological events.

Purpose of the Study:

• To develop and evaluate a novel, low-complexity, real-time image-processing approach for detecting neonatal clonic seizures.
• To assess the algorithm's performance using sensitivity and specificity metrics.

Main Methods:

• Extraction of an average luminance signal from newborn video to represent body movements.
• Utilizing a hybrid autocorrelation-Yin estimation technique to analyze signal periodicity within time windows.
• Processing applied on single and interlaced window bases.

Main Results:

• The proposed method demonstrates the potential for detecting periodic movements characteristic of clonic seizures.
• Performance evaluation using receiver operating characteristic (ROC) curves indicates the algorithm's sensitivity and specificity.
• The approach is validated on video recordings of newborns diagnosed with neonatal seizures.

Conclusions:

• The developed image-processing technique provides a promising, low-complexity, real-time solution for neonatal clonic seizure detection.
• The method's reliance on analyzing movement periodicity from video luminance signals offers a non-invasive diagnostic aid.
• Further validation and clinical integration could enhance neonatal seizure management.