Identify MRI negative temporal lobe epilepsy with resting fMRI indicators and machine learning techniques

Area of Science:

• Neuroimaging
• Machine Learning
• Epilepsy Diagnosis

Background:

• Approximately 30% of temporal lobe epilepsy (TLE) cases are MRI-negative, complicating diagnosis.
• Accurate differentiation of MRI-negative TLE from healthy individuals is clinically significant.

Purpose of the Study:

• To evaluate machine learning (ML) models for diagnosing MRI-negative TLE using resting-state functional MRI (rs-fMRI) metrics.
• To compare the diagnostic performance of single vs. combined rs-fMRI indices.

Main Methods:

• Retrospective analysis of rs-fMRI data from 90 MRI-negative TLE patients and 90 healthy controls.
• Extraction of functional indices: degree centrality (DC), voxel-mirrored homotopic connectivity (VMHC), regional homogeneity (ReHo), fractional amplitude of low-frequency fluctuations (fALFF), and amplitude of low-frequency fluctuations (ALFF).
• Classification using support vector machines (SVM), random forests (RF), and logistic regression (LR) on training/testing sets; performance evaluated by AUC, accuracy, sensitivity, and specificity.

Main Results:

• The SVM model combining rs-fMRI indices achieved the highest performance, with an AUC of 0.89 and 82% accuracy on the test set.
• Amplitude of low-frequency fluctuations (ALFF) in the cerebellum was the most significant feature in the optimal SVM model.
• Models using individual rs-fMRI indices showed lower diagnostic performance, with RF using DC achieving only 47% accuracy.

Conclusions:

• Combined rs-fMRI indices, particularly within an SVM framework, show significant potential for diagnosing MRI-negative TLE.
• rs-fMRI metrics offer a complementary approach for classifying TLE patients when conventional MRI is inconclusive.
• Further validation is warranted to integrate these ML-based rs-fMRI findings into clinical practice.