Performance Analysis of Feature Extraction and Deep Learning Approaches on Whole and Segmented Histopathological Images for Cancer Grade Classification

Breast cancer is a leading cause of cancer-related deaths globally, necessitating effective diagnostic measures. Tissue biopsy examination and histopathology image analysis are pivotal in clinical cancer diagnosis. Various feature extraction approaches and classifiers have been introduced for histopathology analysis by the research community. This study focuses on enhancing tissue classification systems through two distinct pipelines applied to the Databiox dataset. In the first pipeline, histopathological images undergo segmentation using the watershed algorithm. Subsequently, features are extracted from the segmented images using the gray level co-occurrence matrix and VGG16 model. The extracted features are then classified using two classifiers, namely dense layer of CNN and Random Forest. The second pipeline involves extracting features directly from entire histopathological images, with subsequent classification using gray level co-occurrence matrix and VGG16 model. Dense layer and RF classifiers are employed to classify the features from whole images. Additionally, Principal Component Analysis is employed to select crucial feature vectors. The evaluation of direct and segmented histopathological images from the Databiox dataset, with a focus on segmentation, reveals a gap in existing literature. Experimental results demonstrate that the VGG16 model, coupled with a classifier, yields superior outcomes when applied to entire histopathological images compared to segmented images.