Introduction
I am thrilled to announce that my abstract has been accepted for presentation at the prestigious 4th International Congress on Mobile Health and Digital Technology in Epilepsy. This groundbreaking conference, to be held in Lausanne, Switzerland, brings together experts and researchers from around the world to discuss advancements in epilepsy management through the use of mobile health and digital technologies.
Problem
Epilepsy, a neurological disorder affecting millions globally, presents unique challenges in its management, particularly in identifying the epileptogenic zone (EZ), the area responsible for generating seizures. Accurate localization of the EZ is crucial for surgical interventions and improved patient outcomes. However, this task is complex and requires analyzing vast amounts of electroencephalography (EEG) data. To address this challenge, automated machine learning (AutoML) models have emerged as a promising tool for EZ localization.
Methods
In my study, I investigated the feasibility of using publicly available AutoML models for EZ localization and evaluated the impact of different hospital settings on their performance. A total of 53 patients diagnosed with epilepsy who underwent long-term stereo-electroencephalography (SEEG) monitoring at hospitals in Brno and Montreal were selected for this study. The SEEG signals were preprocessed and analyzed using four state-of-the-art AutoML models: H2O.ai, TPOT, Auto-Keras, and Auto-Sklearn. To assess their performance, I employed metrics such as sensitivity, specificity, accuracy, and precision-recall curves.
Results
The results showed that the four AutoML models had varying degrees of accuracy in localizing the EZ. In Brno, Auto-Keras achieved the highest sensitivity of 91.30%, while Auto-Sklearn achieved the highest specificity of 99.27%. TPOT and H2O achieved the highest precision of 70.23% and 63.40%, respectively. In Montreal, TPOT achieved the highest sensitivity of 65.96%, while Auto-Sklearn achieved the highest specificity of 95.24%. TPOT achieved the highest precision of 84.40%, and Auto-Sklearn achieved the highest recall of 30.21%.
Recommendation
The comparison between the hospitals in Brno and Montreal shed light on the impact of different hospital settings on the performance of AutoML models for EZ localization. These findings highlight the need for tailored approaches in optimizing the performance of AutoML models based on specific hospital settings. Additionally, I propose that further clinical research studies are necessary to assess the practical implications of AutoML models in the management and treatment of epilepsy patients.
Attending the Conference
I am immensely grateful for the opportunity to present my research at the 4th International Congress on Mobile Health and Digital Technology in Epilepsy in Lausanne, Switzerland. This conference serves as an ideal platform for sharing knowledge and collaborating with experts in the field. I am excited to connect with researchers, clinicians, and industry professionals who are passionate about leveraging mobile health and digital technologies to advance epilepsy management.
Conclusion
This milestone marks an important step in my academic and research journey. Presenting my findings at the conference will not only allow me to contribute to the collective knowledge in the field but also provide valuable insights from the perspectives of different hospital settings. I am confident that my study's results will contribute to the ongoing efforts in optimizing AutoML-based EZ localization and further enhancing the management and treatment of epilepsy. With enthusiasm, I look forward to attending the conference and engaging in fruitful discussions with the global epilepsy community.
Marek