Identification of Malignancy from Cytological Images Based on Superpixel and Convolutional Neural Networks
This chapter explores two methodologies for classification of cytology images into benign and malignant. Heading toward the automated analysis of the images to eradicate human intervention, this chapter draws curtain from the history of automated CAD-based design system for better understanding of the roots of the evolving image processing techniques in the analysis of biomedical images. Our first approach introduces the clustering-based approach to segment the nucleus region from the rest. After segmentation, nuclei features are extracted based on which classification is done using some standard classifiers. The second perspective suggests the usage of deep-learning-based techniques such as ResNet and InceptionNet-v3. In this case, classification is done with and without segmented images but not using any handcrafted features. The analysis provides results in favor of CNN where the average performances are found better than the existing result using feature-based approach.