Graph based Automatic Breast Tumor Classification Through Ultrasound Images by Radiomics Features

Abstract

Breast tumor a common and devastating cancer in women, poses significant challenges in both diagnosis and prognostication. In this study, we aim to enhance the classification of breast tumor by breast ultrasound image data and a comprehensive set of medical features extracted through the PyFeats framework. Breast tumors, with their varied manifestations and harmful effects, necessitate an advanced approach for accurate classify. This study focuses on the extraction of nine distinct medical features: First Order Statistics, Gray Level Difference Statistics, Statistical Feature Matrix, Gray Level Run Length Matrix, Gray Level Size Zone Matrix, Higher Order Spectra, Local Binary Pattern, Discrete Wavelet Transform, and Stationary Wavelet Transform. These features are meticulously computed based on tumor annotation, offering a detailed characterization of breast tumors and their intricacies. To further improve the accuracy of tumor classification, we employ various machine learning and deep learning algorithms. Notably, we introduce the Radiomic Graph Network model, specifically designed for graph-based data, where nodes symbolize entities and edges signify the intricate relationships between them. The core objective of the GGN model is to generate low-dimensional vector representations (embeddings) for nodes within the graph, preserving the underlying structural and relational information. The innovative GNN model significantly enhances our ability to discern between different tumor tumor classify with greater precision. The novel GNN model mechanism outperformed the existing methods that consider the state of the art in breast tumor based on medical features utilizing 2 stages of breast ultrasound image dataset (BUSI).