[Paper Review] Radiomics in Cancer Radiotherapy: a Review

Jiwoong Jeong, Arif Ali+4 more|arXiv (Cornell University)|Oct 4, 2019

Radiomics and Machine Learning in Medical Imaging134 references4 citations

TL;DR

This review synthesizes the current state of radiomics in cancer radiotherapy, emphasizing its workflow—image acquisition, segmentation, feature extraction, and model validation—while highlighting applications across CT, MRI, PET, and ultrasound. It identifies radiomics as a non-invasive, repeatable method for quantifying tumor phenotypes and predicting treatment response, with key contributions in advancing personalized oncology through machine learning and quantitative imaging biomarkers.

ABSTRACT

Radiomics is a nascent field in quantitative imaging that uses advanced algorithms and considerable computing power to describe tumor phenotypes, monitor treatment response, and assess normal tissue toxicity quantifiably. Remarkable interest has been drawn to the field due to its noninvasive nature and potential for diagnosing and predicting patient prognosis. This review will attempt to comprehensively and critically discuss the various aspects of radiomics including its workflow, applications to different modalities, potential applications in cancer radiotherapy, and limitations.

Motivation & Objective

• To provide a comprehensive and critical overview of radiomics in cancer radiotherapy, focusing on its clinical potential and technical challenges.
• To examine the radiomics workflow across imaging modalities, including image acquisition, segmentation, feature extraction, and model validation.
• To evaluate the role of radiomics in improving tumor characterization, treatment response prediction, and normal tissue toxicity assessment.
• To identify key limitations such as lack of standardization in image acquisition, feature extraction, and nomenclature across institutions.
• To advocate for clinical translation by promoting standardization, data sharing, and validation frameworks like TRIPOD and IBEX.

Proposed method

• Systematic review of radiomics literature from the past decade using PubMed and manual review of cited sources.
• Workflow analysis of radiomics: image acquisition, segmentation of regions of interest (ROIs), feature extraction (shape, texture, histogram, filtered-based), and model development.
• Application of machine learning and data mining techniques to extract high-throughput quantitative features from medical images (CT, MRI, PET, US).
• Evaluation of feature reproducibility and robustness using standardized platforms such as CERR, CGITA, MaZda, and IBEX.
• Assessment of clinical applicability through validation of radiomic models using independent datasets and transparent reporting standards (e.g., TRIPOD).
• Integration of radiomic features with clinical outcomes (e.g., survival, treatment response) to build predictive models for personalized radiotherapy.

Experimental results

Research questions

• RQ1 How can radiomics improve the non-invasive, quantitative characterization of tumor phenotypes in cancer radiotherapy?
• RQ2 What are the key components and challenges in the radiomics workflow across different medical imaging modalities?
• RQ3 To what extent can radiomic features predict treatment response and patient survival in cancers such as glioblastoma, breast, and head and neck cancer?
• RQ4 What are the major barriers to clinical translation of radiomics, and how can standardization address them?
• RQ5 How can radiomics be integrated into routine clinical practice to enhance personalized radiotherapy planning and outcome prediction?

Key findings

• Radiomics enables high-throughput, objective, and quantitative analysis of tumor phenotypes using advanced algorithms on medical images, offering a non-invasive alternative to repeated biopsies.
• Radiomic features extracted from CT, MRI, PET, and ultrasound show strong prognostic value in cancers such as glioblastoma multiforme, breast cancer, and renal cell carcinoma.
• Despite promising results, most studies are retrospective and lack standardization in image acquisition, segmentation, and feature naming, limiting reproducibility and clinical adoption.
• Software platforms like IBEX, CERR, CGITA, and MaZda support feature extraction and data sharing, but universal standardization across institutions remains a critical gap.
• The integration of radiomics into clinical practice depends on three factors: enhancing current predictive models, enabling early treatment response detection, and ensuring robust validation through frameworks like TRIPOD.
• Radiomics adds no additional radiation burden, as it post-processes images already acquired in routine clinical care, making it ideal for repeated monitoring of tumor dynamics.