A0495
Title: Enhancing subarachnoid hemorrhage monitoring with AI and uncertainty analysis
Authors: Robertas Alzbutas - Kaunas University of Technology, Lithuanian Energy Institute (Lithuania) [presenting]
Tomas Iesmantas - Kaunas University of Technology (Lithuania)
Jewel Sengupta - Kaunas University of Technology (Lithuania)
Abstract: The focus is on improving the monitoring of subarachnoid hemorrhage (SAH) through advanced artificial intelligence techniques. Traditional segmentation models are enhanced by integrating image augmentation and adding a regression layer, enabling more precise localization and quantification of SAH from cerebral images. Additionally, a modified region-growing method segments affected brain regions, from which features are extracted using pre-trained models. Dimensionality reduction is applied via optimization algorithms, and an unsupervised deep learning model based on spatial distance is used for automatic SAH segmentation. This model achieves accurate, regular contours quickly. Additionally, optimized feature vectors are classified using an autoencoder to identify SAH subtypes. The approach's effectiveness is validated through case studies, emphasizing feature selection and prediction accuracy, aiding clinicians in making informed treatment decisions. The presented framework integrates AI with uncertainty analysis, combining multiple models to assess prediction performance, analyze result sensitivity, and reduce overall uncertainty through sampling-based methods and tolerance intervals. This methodology facilitates timely data incorporation and improved training processes, enhancing the identification of critical contributors to such SAH monitoring.
