Improving Clinical Accuracy and False Alarm Sensitivity of Tonic-Clonic Seizure Predictors
Session Number
MEDH 02
Advisor(s)
Anthony Cuturrufo, UCLA
Discipline
Medical and Health Sciences
Start Date
17-4-2024 11:05 AM
End Date
17-4-2024 11:20 AM
Abstract
Machine learning applications in diagnosing and predicting seizure symptoms among high-risk patients are critical to prevent life-threatening injuries. Despite existing research achieving near-perfect accuracies in controlled environments, existing models often falter in clinical settings and wearable devices, resulting in high false alarm rates and lower prediction success for Tonic-Clonic seizures. The OpenSeizureDatabase has revolutionized training models by incorporating previously unavailable false-alarm data and prediction failures of seizure predictors. To address these challenges, I conducted a comprehensive data analysis, and using novel machine learning techniques, I have employed a Convolutional Neural Network (CNN) to achieve an 88% predictive accuracy. This new model aims to enhance both false alarm reduction and the detection of previously missed seizures. This research places emphasis on the practicality of deploying machine learning models in their intended use in clinical settings, while representing a significant advancement in improving the performance of seizure prediction models. Shifting focus of diagnostic machine learning models to accuracy in clinical environments is essential for enhancing high-risk patient safety and care.
Improving Clinical Accuracy and False Alarm Sensitivity of Tonic-Clonic Seizure Predictors
Machine learning applications in diagnosing and predicting seizure symptoms among high-risk patients are critical to prevent life-threatening injuries. Despite existing research achieving near-perfect accuracies in controlled environments, existing models often falter in clinical settings and wearable devices, resulting in high false alarm rates and lower prediction success for Tonic-Clonic seizures. The OpenSeizureDatabase has revolutionized training models by incorporating previously unavailable false-alarm data and prediction failures of seizure predictors. To address these challenges, I conducted a comprehensive data analysis, and using novel machine learning techniques, I have employed a Convolutional Neural Network (CNN) to achieve an 88% predictive accuracy. This new model aims to enhance both false alarm reduction and the detection of previously missed seizures. This research places emphasis on the practicality of deploying machine learning models in their intended use in clinical settings, while representing a significant advancement in improving the performance of seizure prediction models. Shifting focus of diagnostic machine learning models to accuracy in clinical environments is essential for enhancing high-risk patient safety and care.