WEDNESDAY, March 19, 2025 (HealthDay News) — A graph neural network using data from the Multicenter Epilepsy Lesion Detection (MELD) Project (MELD Graph) can detect epileptogenic focal cortical dysplasia (FCD) on magnetic resonance imaging (MRI) scans, with improved positive predictive value (PPV) compared with an existing algorithm, according to a study published online Feb. 24 in JAMA Neurology.

Mathilde Ripart, Ph.D., from the UCL Great Ormond Street Institute of Child Health in London, and colleagues examined the efficacy and interpretability of graph neural networks in automatically detecting FCD lesions on MRI scans in a multicenter diagnostic study. Retrospective MRI data were obtained from 23 epilepsy centers worldwide between 2018 and 2022. Data from 20 centers were split into training and testing cohorts, and data from three centers were used for site-independent testing. MELD Graph was trained to identify FCD on surface-based features; performance was compared to that of an existing algorithm. Overall, 34 surface-based MRI features and manual lesion masks were obtained from 703 patients with FCD-related epilepsy and 482 controls.

The researchers found that the MELD Graph had a sensitivity of 81.6 and 63.7 percent in histopathologically confirmed patients who were seizure-free one year after surgery and in MRI-negative patients with FCD, respectively, in the test dataset. The PPV was 67 percent for 260 patients in the test dataset compared with 39 percent using an existing baseline algorithm. In the independent test cohort, the corresponding PPVs were 76 and 46 percent.

“With improved performance over existing methods, interpretable predictions, model confidence scores, and individual patient reports, MELD Graph can support the integration of lesion detection tools into the radiological workflow,” the authors write.

Several authors disclosed ties to the biopharmaceutical industry.

Abstract/Full Text (subscription or payment may be required)

Copyright © 2025 HealthDay. All rights reserved.