Texture features based microscopic image classification of liver cellular granuloma using artificial neural networks

Automated classification of Schistosoma mansoni granulomatous microscopic images of mice liver using Artificial Intelligence (AT) technologies is a key issue for accurate diagnosis and treatment. In this paper, three grey difference statistics-based features, namely three Gray-Level Co-occurrence Matrix (GLCM) based features and fifteen Gray Gradient Co-occurrence Matrix (GGCM) features were calculated by correlative analysis. Ten features were selected for three-level cellular granuloma classification using a Scaled Conjugate Gradient Back-Propagation Neural Network (SCG-BPNN) in the same performance. A cross-entropy is then calculated to evaluate the proposed Sigmoid input and the ten-hidden layer network. The results depicted that SCG-BPNN with texture features performs high recognition rate compared to using morphological features, such as shape, size, contour, thickness and other geometry-based features for the classification. The proposed method also has a high accuracy rate of 87.2% compared to the Back-Propagation Neural Network (BPNN), Back-Propagation Hopfield Neural Network (BPHNN) and Convolutional Neural Network (CNN).