A Fast and Efficient Semi-Unsupervised Segmentation and Feature-Extraction Methodology for Artificial Intelligence and Radiomics Applications: A Preliminary Study Applied to Glioblastoma

Abstract

Brain tumors are pathologies characterized by a high degree of mortality. An early diagnosis of these pathologies could reduce mortality and limit the adverse effects of brain surgery. Computer-aided tomography (CT), and magnetic resonance imaging (MRI) are fundamental diagnostic methods. They offer lots of helpful information that help medical operators to make an early and effective diagnosis. However, a human operator must analyze and classify the enormous amount of data provided. This process is time-consuming, and sometimes the information is not directly visible to the human eye, leading to lost essential information that could be useful for obtaining a correct and early diagnosis. In such a scenario, the development of suitable tools aimed at helping the human operator is essential. In particular, artificial intelligence (AI) methodologies could help the clinical operator correctly classify different tumoral pathologies, suggest more appropriate therapy, and support the surgeon in reducing invasiveness. All AI systems require a so-called training phase and suitable feature identification to work properly. In this work, we propose a tool to speed up brain tumor segmentation and feature extraction. In particular, we focus on Glioblastoma (GBM), a brain tumor characterized by high tissue heterogeneity and difficult segmentation. The method has been assessed by considering an experimental dataset belonging to the Radiomic Laboratory of the University of Trento. The obtained results are encouraging and demonstrate that the proposed method can be very useful to speed up the pathologies segmentation and features extraction compared to other well-known methods.