A Two-Stage Framework for Identifying Robust Radiomic Biomarkers in Prostate MRI: Evidence from the PRODIF Study : Joint Repeatability and Reliability Analysis at the Lesion Level

Read full text Summarise with VIRAT
Authors: Ahmad Ichsan
Cited by: 0
References: 155

Summary

No summary yet. VIRAT can read this paper and write a plain-language summary of its aim, methods, key findings and limitations.

Summarise with VIRAT Free to try with your Science Catalogs account.

Abstract

Prostate cancer is the most prevailing non-cutaneous malignant cancer growths among the male population worldwide, which poses obstacles for early detection and clinical decision- making. Despite extensive adoption of multiparametric magnetic resonance imaging (mpMRI), the shortfalls in reliable and repeatable quantitative imaging biomarkers (QIB) persists as a major barrier to the clinical translation of radiomics-based prostate cancer management. Moreover, there is an increasing necessity for minimally invasive prognostic tools which may consolidate diagnostic workflows and assist clinicians in performing triage towards patients requiring either urgent or targeted treatment. mpMRI has become central to prostate cancer diagnostics; however, variability in the imaging acquisition techniques, reconstruction methods, and quantitative analysis can affect diagnostic performance and reproducibility of mpMRI radiomics as a QIB. This thesis evaluates the classification performance, repeatability and reproducibility of multiple MRI modalities through radiomics-based analysis, with primary focus on diffusion basis spectrum imaging (DBSI) and oscillating gradient spin-echo (OGSE) imaging. The study analyses diffusion MRI data from the PRODIF cohort study deriving 37 patients with histopathologically confirmed prostate cancer following prostatectomy. Imaging was performed using a 3T Ingenuity PET/MR with 32-channel cardiac coil. Statistical analysis was confined to DBSI and OGSE modalities to assess the repeatability and quantitative robustness of advanced diffusion parameterisations in prostate cancer imaging. Radiomic feature analysis included features from first-order moments, Gabor filters, fast Fourier transform features, edge- and corner-based features (2D/3D) and PyRadiomics-derived lesion metrics were conducted using R to assess reliability and repeatability across acquisitions through a two-stage Intraclass Correlations (ICC) and Area Under Curve (AUC) statistical analytical framework. Initial results show that when paired with certain MRI modalities, edge- and corner-derived radiomics and PyRadiomics toolkits show potential as repeatable and reliable biomarkers. Secondary stage framework showed only Scharr and Hessian edge- and corner-derived radiomics persisted in both qib-level candidacy analysis in both repeatability, reliability and precision metrics. Through structured repeatability and reliability analysis of modality-specific MRI parameterisations and radiomic features, this thesis identifies structurally stable descriptors with potential translational relevance, while highlighting that broader multi-centre standardisation remains essential for full QIB validation.

Fields of Study & Subjects

Radiomics and Machine Learning in Medical Imaging MRI in cancer diagnosis Prostate Cancer Diagnosis and Treatment

Publication details

Journal / source
UTUPub (University of Turku)
Publication date
4 March 2026
Language
EN
Document type
Thesis