Pathology grading in retina digital images using student-adjusted empirical mode decomposition and power law statistics

We present a modified two-dimensional empirical mode decomposition method (2D-EMD) for biomedical images using Student's probability density function to reduce outlier effects in the envelope estimation step. An application is made for retina pathology grading of high versus low density blot hemorrhage, and high versus low subretinal exudate. Power law regression estimates of image fractal properties in the Fourier domain are used to characterize each pathology grade, and support vector machines (SVM) are employed for classification. On both grading problems, the experimental results indicated that the proposed method outperforms classical 2D-EMD.