Detection of temporal lobe epilepsy using support vector machines in multi-parametric quantitative MR imaging

Area of Science:

• Neuroimaging
• Medical Diagnostics
• Machine Learning in Medicine

Background:

• Detecting temporal lobe epilepsy (TLE) using MRI is challenging, as many patients lack discernible abnormalities on standard scans.
• Quantitative MRI techniques offer more detailed information than conventional imaging, potentially revealing subtle disease markers.

Purpose of the Study:

• To develop and evaluate a machine learning model for improved TLE detection by integrating quantitative relaxometry and diffusion tensor imaging (DTI) data.
• To optimize classification models by analyzing feature selection, principal component analysis (PCA), and support vector machine (SVM) parameterization.

Main Methods:

• A multi-parametric quantitative MRI approach was used, including T1 map, T2 map, fractional anisotropy, and mean diffusivity.
• Support vector machines (SVM) were employed for classification, with feature selection via ANOVA and dimensionality reduction using PCA.
• Eight distinct classification models were tested on data from 17 TLE patients and 19 controls.

Main Results:

• The combined quantitative MRI and SVM approach achieved an overall classification accuracy of 88.9% for TLE detection.
• Perfect classification (100% accuracy) was achieved for patients with left-sided seizure onset, highlighting the model's sensitivity to specific disease lateralization.
• The study identified a linear SVM combined with ANOVA feature selection and PCA as an effective strategy for high-dimensional data with limited sample sizes.

Conclusions:

• Multi-parametric quantitative MRI combined with ROI-based SVM classification is a promising tool for identifying TLE patients, particularly those with non-obvious lesions on standard MRI.
• The method has the potential to enhance diagnostic assessment and improve patient management in TLE.
• Disease heterogeneity impacts classification outcomes, suggesting that patient stratification may further refine diagnostic accuracy.