Exploring Radiomics and Unveiling Novel Qualitative Imaging Biomarkers for Glioma Diagnosis in Dogs

Radiomics integrates machine learning (ML) and radiology to extract and analyze quantitative features from medical imaging modalities such as Magnetic Resonance Imaging (MRI), Computed Tomography (CT), Positron Emission Tomography (PET), ultrasound (US) and digital radiographs (DX). By extracting pixel/voxel-level data, followed by standardization and feature selection, radiomics enables ML algorithms to assist in diagnosis and prognosis. While extensively researched in human medicine its application in veterinary medicine remains limited. Radiomics offers objective, data-driven insights, surpassing qualitative evaluations by revealing micromolecular disease features invisible to the human eye. Radiomics holds significant promise for diagnosing gliomas (GM), a challenging brain tumor where histopathology, the diagnostic gold standard, is seldom performed in veterinary medicine due to logistical and financial barriers, and it is also limited by inherent pathologist subjectivity and disagreement. Additionally, qualitative MRI demonstrates limited accuracy in identifying GM type and grade. By offering non-invasive and reproducible diagnostic and prognostic solutions, radiomics has the potential to overcome these challenges, enhancing brain tumor evaluation in both veterinary and human medicine. The primary goal of this study is to enhance the diagnosis and prognosis of GM by exploring both conventional and innovative non-invasive imaging techniques, with a focus on qualitative and quantitative MRI approaches. We hypothesize that quantitative and novel qualitative methods will surpass conventional expert qualitative assessments in accurately diagnosing GM type, grade, and progression. By doing so, we aim to improve the precision of GM imaging diagnoses, offering clinicians a more accessible and reliable tool to support their diagnostic and treatment decisions. Chapter 1 of this dissertation presents a comprehensive review of the challenges associated with diagnosing GM using MRI. It also introduces principles of radiomics, a novel and relatively underexplored field in veterinary medicine centered on quantitative imaging analysis for diagnostic and prognostic purposes. This includes an in-depth discussion of the radiomics workflow and associated ML methods. Chapter 2 demonstrates the use and efficacy of quantitative MRI for determination of GM size and therapeutic response assessments using both linear and volumetric techniques. Chapter 3 investigates the T2-weighted–FLAIR mismatch sign (T2FMM) in dogs, a well-established imaging biomarker of human low-grade astrocytomas, and demonstrates that the T2FMM is a highly specific biomarker for oligodendrogliomas —the first such imaging biomarker for GM to be discovered in veterinary research. Finally, Chapter 4 illustrates a structured radiomics pipeline for the standardized quantitative analysis of brain tumors on MRI and demonstrates that the use of radiomics ML models results in superior ability to diagnose canine GM subtypes and grades and discriminate GM from non-neoplastic intra-axial lesions when compared to expert rater opinions derived from qualitative MRI evaluations.