A Machine Learning Framework for Automatic and Continuous MMN Detection With Preliminary Results for Coma Outcome...

Area of Science:

• Neuroscience
• Computational Neuroscience
• Medical Technology

Background:

• Mismatch negativity (MMN) is a crucial auditory event-related potential (ERP) component linked to coma emergence.
• Traditional MMN detection via visual inspection is subjective, time-consuming, and often impractical.
• Accurate MMN detection is vital for predicting coma recovery.

Purpose of the Study:

• To develop and validate a practical machine learning (ML) based approach for automated MMN detection.
• To enhance the accuracy and efficiency of predicting emergence from coma.
• To enable continuous monitoring of conscious state fluctuations with high temporal resolution.

Main Methods:

• An auditory odd-ball paradigm was used with 22 healthy subjects and 2 coma patients.
• A novel ML approach was developed for automatic MMN detection, utilizing similarity measures against healthy subject responses.
• The method was validated for MMN detection accuracy and prediction of coma emergence.

Main Results:

• The ML method achieved 92.7% accuracy in MMN detection in healthy subjects.
• The approach successfully predicted coma emergence in both patients, outperforming conventional methods.
• The ML model demonstrated capability for MMN detection in short intervals (as brief as two minutes).

Conclusions:

• The proposed ML-based MMN detection offers a practical, accurate, and automated solution for predicting coma emergence.
• Training solely on healthy subjects presents a novel and efficient approach for ML model development in clinical settings.
• This method holds potential for continuous monitoring and understanding of consciousness states in critically ill patients.