Malformations of cortical development (MCDs) encompass a variety of brain disorders affecting the normal development and organization of the brain cortex. The relatively low incidence and the extreme heterogeneity of these disorders hamper the application of classical group level approaches for the detection of lesions. Here, we present a geometrical descriptor for a voxel level analysis based on fractal geometry, then define two similarity measures to detect the lesions at single subject level. The pipeline was applied to 15 normal children and nine pediatric patients affected by MCDs following two criteria, maximum accuracy (WACC) and minimization of false positives (FPR), and proved that our lesion detection algorithm is able to detect and locate abnormalities of the brain cortex with high specificity (WACC = 85%, FPR = 96%), sensitivity (WACC = 83%, FPR = 63%) and accuracy (WACC = 85%, FPR = 90%). The combination of global and local features proves to be effective, making the algorithm suitable for the detection of both focal and diffused malformations. Compared to other existing algorithms, this method shows higher accuracy and sensitivity.