Rotational Invariant G-L Fractional Derivative Filters for Lung Tissue Classification

Abstract

We propose a new and powerful rotation invariant Convolution Neural Network (G-LCNN) model for the classification of five categories of Interstitial Lung Diseases (ILD) such as Normal, Emphysema, Ground Glass, Fibrosis, and Micronodules. Grunwald-Letnikov (G-L) Fractional derivative filters find applications in the field of medical image analysis since they enhance homogenous areas and tiny edges. These filters are applied to the lung image patch in eight directions to have rotation-invariant features that are helpful to solve the pose variations of the patient caused during lung CT scanning. To evaluate the G-LCNN model publicly available Interstitial Lung Disease (ILD) database with 24,105 image patches is considered and the results show superior performance with an average F-Score and Accuracy noted as 97.01% and 96.33% respectively.