Epilepsy is a debilitating disorder that affects 50 million people worldwide, including one in 150 children. Some patients with epilepsy (including roughly 20-40% of children) suffer from refractory seizures despite treatment with anticonvulsants, i.e. pharmaco-resistant epilepsy (PRE), which results in increased morbidity and mortality. Early identification and surgical intervention of PRE, not only correlates to better outcomes, but is crucial to avoid complications which could cause sudden death. Surgical intervention, however, requires precise SOZ localization for resection/disconnection or ablation. The current gold standard for SOZ localization uses invasive intracranial electroencephalography (iEEG), which requires implanting depth electrodes guided by an initial SOZ localization. A couple alternatives that have been explored are 1. manually guiding the iEEG lead placement to the expected SOZ location and 2. bypassing iEEG monitoring using DL on brain images from resting-state functional MRI (rs-fMRI) and diffusion MRI (dMRI). However, these approaches exhibit poor precision and may not be feasible in a clinical setting.
Researchers at Arizona State University have developed a novel human-AI collaboration that identifies SOZ in focal epilepsy patients called DeepXSOZ. This collaboration uses entropy imbalance gain and Gini index to quantify class imbalance and intra-class variability. It also orchestrates supervised AI and expert knowledge machines to effectively identify rare class through human-AI collaboration with reduced human effort. DeepXSOZ is able to overcome the performance drawbacks of many SOZ localization strategies. When tested on 52 children with PRE, DeepXSOZ, compared to state-of-the art, consistently maintains a statistically stable and higher accuracy, prevision and sensitivity across all the age groups and sex distribution.
DeepXSOZ allows the surgical team to evaluate the automation fidelity and choose the appropriate level of automation and manual effort for optimal patient outcome.
Potential Applications
- Could enable the usage of rs-fMRI as a low-cost outpatient presurgical screening tool to identify SOZ
Benefits and Advantages
- Automated identification of SOZ localizing ICs which are relatively infrequent in a dataset
- Reduces expert sorting workload by 7-fold
- Reduced costs and time
- Enables the usage of rs-fMRI as a low-cost outpatient pre-surgical screening tool
- Mitigates class imbalance and intraclass variability effects
- May reduce false positives and increase true positives of SOZ localizing ICs
- Minimal data leakage effect with statistically similar performance across multi-center datasets without fine tuning
- Comparison with state-of-the art, on 52 children with PRE, shows a sensitivity of 95.4%, precision of 91.3% and accuracy of 87.5%
- Achieves significantly higher and consistent results across age and sex
- the time commitment for presurgical evaluation
For more information about this opportunity, please see
For more information about the inventor(s) and their research, please see