ENGINEERING FOR HEALTH - Evaluation of Radiomic Analysis over the Comparison of Machine Learning Approach and Radiomic Risk Score on Glioblastoma

Aug 6, 2026

Upload by Finn

Accurate prognosis in glioblastoma (GBM) supports effective treatment planning, and radiomics analysis provides quantitative image-derived features for diagnosis, prognosis, and therapy response modeling. This research extracts standardized radiomic features from GBM MRI scans, performs feature selection, and compares radiomic-based risk scores (RRS) with machine learning (ML) approaches for patient risk stratification. Model generalisability is evaluated for clinical translation. Results show a logistic regression stratification model generalises better than the RRS method on unseen datasets.

LanguageEnglish

CategoryResearch & Report

FormatPDF

Pages5

Reading Time13 min

AccessFree to download

Introduction

Background and clinical need

Radiomics analysis and modelling

Prior work and research aim

Materials and Methods

Datasets and MRI sequences

Training/testing strategy

Image preprocessing and segmentation

Duman A Cardiff University School of Engineering Powell J Velindre NHS Trust, Department of Oncology, Cardiff Thomas S Velindre NHS Trust, Department of Oncology, Cardiff Spezi E Cardiff University School of Engineering ENGINEERING FOR HEALTH Evaluation of Radiomic Analysis over the Comparison of Machine Learning Approach and Radiomic Risk Score on Glioblastoma Accurate patient prognosis is important to provide an effective treatment plan for Glioblastoma (GBM) patients. Radiomics analysis extracts quantitative features from medical images. Such features can be used to build models to support medical decisions for diagnosis, prognosis, and therapeutic response. The progress of radiomics analysis is continuously improving. The aim of this research is to extract standardised radiomic features from MRI scans of GBM patients, perform feature selection, and compare radiomicbased risk score (RRS) and machine learning (ML) approaches for the risk stratification of GBM patients. We have also tested the generalisability of these models which is crucial for clinical implementation. Our work demonstrates that a stratification model based on logistic regression generalised better than the RRS method when applied to new unseen datasets. Keywords: Glioblastoma, radiomics, brain tumour, overall survival. Corresponding author: [DumanA@cardiff.ac.uk](DumanA@cardiff.ac.uk) A. Duman, J. Powell, S. Thomas, and E. Spezi,‘Evaluation of Radiomic Analysis over the Comparison of Machine Learning Approach and Radiomic Risk Score on Glioblastoma’, Proceedings of the Cardiff University Engineering Research Conference 2023, Cardiff, UK, pp. 19-22. [doi.org/10.18573/conf1.f](doi.org/10.18573/conf1.f) 20 Proceedings of the Cardiff University Engineering Research Conference 2023 INTRODUCTION Glioblastoma (GBM) is a malignant and lethal brain tumour [1] . Grade IV gliomas exhibit the highest level of aggression and rapid progression. After initial diagnosis, the median survival time for GBM patients is 15 months [2] . The poor prognosis for GBM can be related to genetic heterogeneity between patients and at intratumor level [3] . In clinical practice, brain tumours are evaluated for their diagnosis and prognosis by utilising magnetic resonance imaging (MRI) techniques. The location of tumours is detected in three dimensions via non-invasive MRI technology. In contrast to X-ray and CT imaging, MRI gives high resolution with better soft tissue contrast without the use of ionising radiation [4] . Biopsies are an invasive procedure to diagnose, grade and characterise brain tumours [5]. Due to having genetic differences in sub-regions of a tumour, biopsies can provide only limited information with a sample of small section from tumour tissue [6]. Other assessment methods including quantitative image analysis, which is non-invasive and evaluate the entire tumour tissue, can support biopsy as additional assessment. Image analysis utilising radiomics features has the potential to replace biopsies when they are infeasible or risky [7] . Radiomics analysis is a rapidly growing field of medical imaging involving the extraction of large amounts of quantitative data from medical images [8],[9] . This approach seeks to reveal hidden patterns and features that are imperceptible by the naked eye in order to provide patients with more personalised and precise care. To extract radiomic features from medical images, radiomics analysis employs advanced image processing techniques that can characterise tumour heterogeneity [10] and microenvironment [11]. Radiomic imaging features can be then used to train a model to stratify patients in different risk groups. This is achieved using statistical methods and machine learning (ML) techniques as outlined in the literature [9] . N. Beig et al. proposed a Radiomics-based Risk Score (RRS) for GBM tumour habitat [12] . However, a comparison with alternative methods including machine learning

FAQs

How does the proposed ML stratification model perform compared with the radiomic risk score (RRS)?

A stratification model based on logistic regression generalises better than the RRS method when applied to new, unseen datasets, supporting improved generalisability.

Aug 6, 2026

Upload by Finn