Performance metrics for the accurate characterisation of interictal spike detection algorithms

Area of Science:

• Neuroscience
• Biomedical Engineering
• Signal Processing

Background:

• Automated detection of epileptic spikes in electroencephalography (EEG) is crucial for efficient and accurate analysis.
• Evaluating the performance of these automated algorithms is challenging due to numerous variables in recordings and algorithms.

Purpose of the Study:

• To summarize key variables affecting automated epileptic EEG spike detection performance evaluation.
• To present and compare different methods for calculating average algorithm performance, including novel weighting factors.

Main Methods:

• Identification and summarization of core variables influencing EEG spike detection performance.
• Development and application of various performance calculation methods incorporating weighting factors.
• Analysis of the impact of four distinct weighting factors on algorithm performance metrics.

Main Results:

• Weighting factors significantly influence the apparent performance of automated spike detection algorithms.
• Non-ideal test cases require correction using appropriate weighting factors for accurate evaluation.
• The 'duration divided by the number of events' weighting factor demonstrated a notable effect.

Conclusions:

• Standardized performance evaluation is essential for automated epileptic EEG spike detection algorithms.
• The proposed weighting factors offer a more accurate assessment of algorithm performance.
• Recommends the 'duration divided by the number of events' weighting for future studies to ensure reliable and comparable results.