Automatic localization of cerebral cortical malformations using fractal analysis

Area of Science:

• Neuroimaging
• Medical Image Analysis
• Computational Neuroscience

Background:

• Malformations of cortical development (MCDs) are diverse brain disorders impacting cortical organization.
• The low incidence and heterogeneity of MCDs challenge traditional group-level lesion detection methods.
• Accurate single-subject analysis is crucial for diagnosing and understanding MCDs.

Purpose of the Study:

• To develop and validate a novel voxel-level lesion detection algorithm for MCDs.
• To enable accurate identification of brain cortex abnormalities at the individual patient level.
• To improve diagnostic capabilities for rare and heterogeneous neurological disorders.

Main Methods:

• A geometrical descriptor based on fractal geometry was developed for voxel-level analysis.
• Two similarity measures were defined to detect lesions at the single-subject level.
• The algorithm was tested on pediatric patients with MCDs and healthy controls, optimizing for accuracy and minimizing false positives.

Main Results:

• The algorithm demonstrated high specificity (96%), sensitivity (63%), and accuracy (90%) in detecting MCD lesions.
• It achieved a weighted accuracy (WACC) of 85% for specificity, 83% for sensitivity, and 85% for overall accuracy.
• The method effectively detected both focal and diffused malformations, outperforming existing algorithms in accuracy and sensitivity.

Conclusions:

• The proposed fractal geometry-based algorithm provides a robust and accurate method for single-subject lesion detection in MCDs.
• This approach overcomes limitations of traditional methods, offering improved diagnostic potential for heterogeneous brain malformations.
• The combination of global and local features makes the algorithm versatile for various types of cortical abnormalities.