Support Vector Machine with nonlinear-kernel optimization for lateralization of epileptogenic hippocampus in MR...

Area of Science:

• Medical Imaging
• Neurology
• Artificial Intelligence

Background:

• Medically intractable epilepsy often requires surgical intervention for seizure control.
• Accurate pre-surgical evaluation, including epilepsy lateralization, is crucial for successful surgical outcomes.
• Magnetic Resonance Imaging (MRI) plays a key role in pre-surgical assessment.

Purpose of the Study:

• To propose an optimized pattern recognition approach for lateralizing mesial Temporal Lobe Epilepsy (mTLE) patients.
• To utilize the asymmetry of hippocampal imaging indices from MRI for improved lateralization.
• To enhance the accuracy of pre-surgical evaluation for epilepsy surgery.

Main Methods:

• Utilized T1-weighted and Fluid-Attenuated Inversion Recovery (FLAIR) MRI images from 76 symptomatic mTLE patients.
• Segmented hippocampi automatically and manually, extracting volumetric and intensity features.
• Employed a nonlinear Support Vector Machine (SVM) with an optimized Gaussian Radial Basis Function (GRBF) kernel for feature classification.

Main Results:

• Achieved an 82% correct lateralization rate using leave-one-out cross-validation.
• Demonstrated a high probability of detection for the left side (0.90) with a low false alarm rate (0.04).
• Showed a probability of detection for the right side (0.69) with zero false alarm probability, outperforming other methods.

Conclusions:

• The proposed optimized pattern recognition method offers a valuable tool for pre-surgical evaluation in mTLE patients.
• This MRI-based lateralization technique improves accuracy and reduces false positives compared to existing methods.
• The findings support the integration of this advanced technique into the pre-surgical planning for epilepsy treatment.