Radiogenomics: Combining Medical Imaging and Genetic Information

Innovation in Genetics and Radiology

What Are Omics?

Omics is a general term for scientific fields that comprehensively study different aspects of biological systems. It includes genomics, proteomics, and metabolomics, analyzing biomolecular structures, functions, and interactions.

What Is Radiomics?

Radiomics is the process of extracting large numbers of quantitative features from medical imaging data and analyzing them. It helps improve understanding of tumor heterogeneity and characteristics.

What Is Genomics?

Genomics studies the structure, function, and interactions of an organism’s entire genetic material, its genome. Genomic research aims to understand how genes work, their disease associations, and the health effects of individual genetic variation.

What Is Radiogenomics?

Radiogenomics integrates medical imaging data with genomic information and examines their relationships. This approach is used to develop personalized cancer treatments.

Could Radiogenomics Be the Key to Personalized Medicine?

Radiogenomics focuses on relationships between genomes and imaging phenotypes and is widely applied to address tumor heterogeneity and predict immune responses and disease progression. It is an inevitable consequence of current personalized medicine trends because it costs less than conventional genetic sequencing and provides access to the whole tumor rather than limited biopsy samples. By providing genetic information through voxels (volumetric pixels), radiogenomics may enable personalized treatment targeting a heterogeneous tumor or group of tumors. Beyond quantitative assessment of lesion characteristics, it can distinguish benign from malignant findings and patient characteristics, improving risk stratification for more precise imaging and screening. This article describes its applications in oncology diagnosis, treatment planning, and assessment through a multiomics approach. We also discuss challenges in radiogenomics and the scope and clinical applicability of these methods.

How the Genomic Revolution Changed Medicine’s Outlook

Since the genomic revolution of the early 1990s, cancer research has sought to enable personalized treatments by investigating disease causes at the genetic level. The Human Genome Project was followed, after its completion, by a shift toward more functional genomics, focusing on gene and protein expression profiles and roles. Understanding cancer genetics has changed how we think about and treat the disease. The Cancer Genome Atlas (TCGA) and similar projects have used many samples across numerous cancers to generate large-scale transcriptomic, epigenomic, and proteomic data. However, conventional genetic analysis relies on invasive biopsies or postoperative pathological tissue. These procedures involve risks and possible complications and cannot therefore be applied to every cancer patient.

Tumor Heterogeneity Must Be Considered

Because of intratumoral or intertumoral heterogeneity, tissue biopsies may fail to identify important genetic changes accurately. Samples often come from a small part of a heterogeneous lesion and may not accurately represent its anatomical, functional, or physiological characteristics. More importantly, repeated tissue sampling during treatment to assess response is not possible. Integrating genomic or proteomic profiling into general clinical practice therefore remains difficult.

The Emergence of Radiomic Solutions

Medical imaging is vital in medical science and clinical practice. Its role, however, is rapidly evolving from a major diagnostic tool toward a dominant role in personalized medicine. Conventional assessment uses qualitative features such as tumor density, enhancement patterns, boundary regularity, cellular and cell-free components, anatomical relationships with surrounding structures, and anatomical changes. In contrast, rapidly developing radiomics makes it possible to digitally decode quantitative information from four-channel images, including intensity, texture, shape, and size metrics.

Introduced in 2012, radiomics aims to characterize and predict tumor heterogeneity by extracting quantitative imaging features with high-throughput algorithms, then filtering, clustering, and analyzing the data. Despite advances in transforming digital medical images into various relevant data, major obstacles remain in determining the biological interpretation of radiomic features.

Integrating Different Data Types

As we enter the next era of big data and personalized medicine, many experts have proposed “radiogenomics,” focusing on multiscale links between medical imaging and genomic data. It can be viewed as a combination of radiomics and genomics that addresses the limitations above. It has developed substantially over the past decade, showing potential for noninvasive prognostic and diagnostic approaches by linking quantitative imaging features to genomic signatures, particularly to identify cancer therapy biomarkers. With improved molecular characterization of cancers and advances in texture analysis and machine learning, cancer diagnosis is poised to move toward personalized medicine through radiogenomics.

Since the genomic revolution of the early 1990s, cancer research has focused on investigating disease causes at the genetic level. Completion of the Human Genome Project moved genomics toward a more functional level, emphasizing gene and protein expression profiles and roles. Understanding cancer genetics has changed how we think about and treat the disease. Projects such as The Cancer Genome Atlas (TCGA) have used many samples across numerous cancers to generate large-scale transcriptomic, epigenomic, and proteomic data. However, conventional genetic analysis relies on invasive biopsies or postoperative pathological tissue. These procedures involve risks and possible complications and cannot therefore be applied to every cancer patient. Intratumoral or intertumoral heterogeneity may prevent tissue biopsies from identifying important genetic changes accurately. Samples often come from a small part of a heterogeneous lesion and may not accurately represent its anatomical, functional, or physiological characteristics. More importantly, repeated tissue sampling during treatment to assess response is not possible. Integrating genomic or proteomic profiling into general clinical practice therefore remains difficult.

Medical imaging is vital in medical science and clinical practice. Its role, however, is rapidly evolving from a major diagnostic tool toward a dominant role in personalized medicine. Conventional assessment uses qualitative features such as tumor density, enhancement patterns, boundary regularity, cellular and cell-free components, anatomical relationships with surrounding structures, and anatomical changes. In contrast, rapidly developing radiomics makes it possible to digitally decode quantitative information from four-channel images, including intensity, texture, shape, and size metrics. Introduced in 2012, radiomics aims to characterize and predict tumor heterogeneity by extracting quantitative imaging features with high-throughput algorithms, then filtering, clustering, and analyzing the data. Despite advances in transforming digital medical images into various relevant data, major obstacles remain in determining the biological interpretation of radiomic features.

As we enter the next era of big data and personalized medicine, many experts have proposed “radiogenomics,” focusing on multiscale links between medical imaging and genomic data. It can be viewed as a combination of radiomics and genomics that addresses the limitations above. It has developed substantially over the past decade, showing potential for noninvasive prognostic and diagnostic approaches by linking quantitative imaging features to genomic signatures, particularly to identify cancer therapy biomarkers. With improved molecular characterization of cancers and advances in texture analysis and machine learning, cancer diagnosis is poised to move toward personalized medicine through radiogenomics.

An Overview of Radiogenomic Workflows

The central idea is to combine imaging data, which quantitatively reflect disease phenotypes, with genomic data reflecting molecular activity. Establishing links between genomic and imaging features, then analyzing and distilling this information, provides insight into disease genetics and development. These tools and methods are described below. In 2012, Lambin and colleagues formally introduced imaging genomics, transforming medical images into large amounts of feature information through automated, high-throughput extraction to explore their biological nature and support clinical decisions. The radiogenomic workflow consists of: (i) image acquisition, similar to “case enrollment” in clinical trials, using radiography, ultrasound, magnetic resonance imaging (MRI), positron emission tomography (PET)/CT, and other modalities; (ii) image segmentation, including manual, semiautomatic, and fully automatic approaches; (iii) feature extraction and characterization, obtaining high-dimensional data to describe regions of interest quantitatively, the core of image histology; (iv) feature selection and model development; and (v) clinical applications, the main uses of image histology, including tumor classification, staging, and prognosis prediction (Figure 1).

An Overview of Radiomic–Genomic Relationships Across Omics Fields

Recent developments in radiogenomic technologies have shifted paradigms in investigating radiomics’ clinical applications from a multiomics perspective. Fathi and colleagues combined multiple omics data, including clinical information, MGMT methylation, radiomics, and genetics, to model outcomes accurately in newly diagnosed GBM patients. This research characterized radiogenomics’ precision medicine applications through multiomics for accurate risk-group classification. We therefore divide radiogenomics into three main components in this review. We searched PubMed and Web of Science for relevant articles published between 2017 and 2022 with an impact factor (IF) above 5.

Using Radiogenomic Models to Predict Gene Mutations

Identifying gene mutations is important for cancer diagnosis, treatment selection, and monitoring effectiveness. The most common detection method is tissue biopsy, which is invasive, costly, and time-consuming and thus unsuitable for some patients. Radiomics, meanwhile, automatically extracts and analyzes large numbers of advanced quantitative features from different imaging modalities using standard radiological scans. Combining radiomics and genomics has produced radiogenomics, an emerging field facilitating precision medicine. Further study of radiomic mutation prediction is therefore needed to support more effective cancer detection. This section briefly reviews current knowledge of its role in predicting mutations in brain, lung, colorectal, breast, and kidney tumors. Most studies focus on gene mutations including single-nucleotide changes and insertions/deletions (indels).

Single-nucleotide polymorphisms (SNPs) are the most widely studied genomic variants. They are single-base-pair changes that can affect coding and noncoding DNA and protein expression. A genome-wide association study (GWAS) uses millions of SNPs for molecular genetic analysis, treating them as biomarkers to compare and identify variants influencing complex traits. Using radiogenomics, Seibold and colleagues evaluated the ten most significant SNPs in four genes in 1883 breast cancer patients during the replication stage, then validated them in 753 patients. They found that rs2682585 was strongly associated with late skin toxicity and overall toxic response. 

Similarly, Kerns and colleagues found that AGT, COG2, CAPN9, ARV1, AL512328.1, and LOC101927604 may be associated with progressive radiotoxic hematuria after prostate cancer radiation treatment.

Genotyping is vital to guide treatment because specific therapies are available for genomic subgroups of non-small cell lung cancer (NSCLC). Radiogenomics may provide an important tool for rapid, noninvasive tumor genotyping. Epidermal growth factor receptor (EGFR) mutations and anaplastic lymphoma kinase (ALK) rearrangements are important biomarkers for treating NSCLC with tyrosine kinase inhibitors (TKIs). Lv and colleagues confirmed that lower pSUVmax was linked to mutant EGFR status and, with other clinical factors, could improve discrimination of EGFR mutation status in some NSCLC patients unable to undergo EGFR testing. Their retrospective study included 849 NSCLC patients with EGFR or ALK alterations. Zhang and colleagues extracted 47 PET and 45 CT imaging features using LIFEx and combined them with clinical variables to create a model with high EGFR predictive power. Similarly, Zhang and colleagues predicted EGFR mutations in NSCLC using 485 quantitative texture features from CT images in a training group, showing excellent performance of a model based on image histological features. Nair and colleagues developed an imaging-genomic model identifying EGFR mutations using CT and 18F-FDG-PET-CT. The FDG-PET model distinguished wild-type from mutant EGFR with good performance, an AUC of 0.8. Overall, FGD-PET and CT imaging perform strongly in identifying EGFR mutations and ALK rearrangements in NSCLC tumors.

The growing incidence, morbidity, and mortality of colorectal cancer (CRC) present a major therapeutic challenge. Imaging-genomic studies of KRAS mutation status in CRC have attracted substantial attention. Shin and colleagues observed associations between KRAS mutations and polyp morphology, increased axial length, increased axial-to-longitudinal ratio, and N2 lymph node status. A subsequent study by Lubner and colleagues examined 77 CRC liver metastases, finding correlations between texture parameters and tumor grade, serum carcinoembryonic antigen (CEA), and KRAS mutation status. Yang and colleagues modeled and analyzed relationships between mutations, clinical background, tumor stage, and histological differentiation in 117 CRC patients. Their findings linked image histological features to KRAS/NRAS/BRAF mutations, suggesting that CT could contribute to CRC tumor genotyping and facilitate more precise treatment.

These studies show that conventional imaging and image histological features can predict mutations well in several cancers. Although they cannot be replaced with pancreatic biopsy, their noninvasive and repeatable nature may help clinical treatment decisions. Clinical applications may nevertheless be limited because many radiosensitive SNPs remain undiscovered.

Radiomic Applications in Transcriptomics

Understanding the diverse roles of genes requires transcriptomics: patterns of gene expression at the genetic transcription level in an organism or particular cells under specific conditions. Transcriptomics can clarify differences in gene expression between healthy people and patients. By examining which genes are switched on or off, and in which groups, researchers gain clues about how expression differences affect disease progression. Further investigation of transcriptomic biomarkers is therefore needed to guide cancer detection more accurately. This section briefly reviews their role in brain and kidney tumors.

Some common mutations have unknown clinical significance, limiting their use as biomarkers. Transcriptional biomarkers are superior ways to divide ccRCC into clinically meaningful molecular subgroups. Brooks and colleagues developed a 34-gene expression signature to classify localized ccRCC as high or low risk. Jamshidi and colleagues also applied GSEA to identify oncogenic pathways using MR imaging features. Correlations among 34 gene loci showed concordant variation in gene dosage and mRNA expression, creating an MR imaging, mRNA, and CNV radiogenomic association map for GBM. Another study investigated transcriptomics in hepatocellular carcinoma by linking molecular features with 18F-fluorocholine PET/CT phenotypes and examining possible survival associations. Similarly, Yamamoto and colleagues created a radiogenomic map linking breast cancer MR phenotypes to global gene expression patterns and identified relationships with gene sets of interest. They also found associations between MR image edge-enhancement features and eight lncRNAs, including HOX transcript antisense RNA (HOTAIR).

Combining Radiomics and Epigenetics

How epigenetic changes drive cancer development is an area of intense research interest. Unlike mutations, epigenetic modifications are reversible, making them attractive intervention targets. Cancer is initiated and develops through epigenetic mechanisms, while epigenetic diversity promotes dynamic expression patterns that support tumor evolution and adaptation. By offering new approaches to important relationships between epigenetic and clinicopathological features, radiogenomics highlights the potential of epigenetic marks as personalized medicine biomarkers.

Cen and colleagues developed a clear cell renal carcinoma model to investigate relationships between CT features, RUNX3 methylation, and survival. Indistinct tumor boundaries, left-sided tumors, and intratumoral vessels significantly predicted increased RUNX3 methylation. Kanas and colleagues also demonstrated an association between standard preoperative MRI variables and MGMT methylation in glioblastoma.

Better understanding of epigenetic mechanisms and interactions between epigenomic data and radiomics may provide new insights for radiogenomic strategies.

Clinical Applications

Radiogenomic research has progressed considerably since the start of this decade, highlighting its potential to advance clinical care. As the field remains in its infancy, its full clinical potential has not yet been realized. Nevertheless, many studies show encouraging early clinical applications. We searched PubMed and Web of Science for relevant articles from 2017 to 2022 with an impact factor (IF) above 5.

Relationships Between Imaging and Molecular Phenotypes

Obtaining an accurate molecular phenotype is a prerequisite for targeted therapy. Because malignant tumors are heterogeneous, small biopsy samples may not accurately reflect their mutations. Radiogenomics aims to improve this by revealing relationships between imaging features (imaging phenotypes) and molecular tumor markers (molecular phenotypes).

Detecting IDH1 mutations has major diagnostic and prognostic importance in glioma. GBM can be classified as IDH-wild-type or IDH-mutant according to isocitrate dehydrogenase (IDH) status. Most research predicts IDH status using MR imaging, with modest success. Current work applies machine learning with clinical factors and multimodal MRI features to predict IDH genotypes in high-grade gliomas. Zhang and colleagues, for example, developed a random forest classifier using clinical data and multimodal preoperative features. Chang and colleagues also applied deep learning to predict IDH genotype noninvasively in grade II-IV gliomas using conventional MR imaging. The model may complement invasive sampling with a noninvasive tool, facilitating earlier management and follow-up.

Breast cancer molecular subtypes — luminal A, luminal B, HER2-enriched, and basal-like (Figure 2) — were first proposed by Perou and colleagues to explain differences in treatment response and outcomes. Subsequent studies confirmed their clinical importance and that chemotherapy, endocrine therapy, and HER2-targeted treatment decisions partly align with them. Mazurowski and colleagues examined 48 patients from the cancer imaging archive and extracted 23 DCE-MRI features, linking dynamic enhancement to luminal B breast cancer. This subtype tends to have a higher lesion-to-background parenchymal enhancement ratio. Leithner and colleagues found that DWI features, such as first-order histograms and gray-level co-occurrence matrices, could more accurately identify receptor status and molecular subtypes for luminal B and HER2-enriched cancers. Xie and colleagues distinguished triple-negative disease from other subtypes using multiplanar MR imaging and tumor histogram analysis, with good accuracy against luminal A and HER2-enriched subtypes. Dilorenzo and colleagues investigated MRI background parenchymal enhancement (BPE) for subtype discrimination, finding that mild BPE indicated luminal B or HER2-negative subtypes, whereas marked BPE indicated triple-negative cancer.

New biological insights have defined four molecular medulloblastoma (MB) subgroups: sonic hedgehog (SHH), wingless (WNT), group 3, and group 4. Conventional features from preoperative multiparametric MRI have been associated with these subgroups, enabling subgroup-specific nomograms with varying predictive accuracy. Nomograms based on preoperative multiparametric MRI can reliably predict SHH and group 4 medulloblastoma. Yan and colleagues developed machine learning models to predict MB molecular subpopulations, showing potential for noninvasive prediction.

Using Radiogenomics to Assess Cancer Treatment Effectiveness and Select Treatment Options

Recent years have brought unprecedented progress in radiogenomics, showing that tumor imaging features can precisely predict responses to treatments including chemotherapy, radiotherapy, targeted therapy, and immunotherapy.

Immunotherapy

18F-FDG-PET/CT imaging genomics is increasingly studied and applied to predict PD-1/PD-L1 expression. Recent studies suggest that a deep-learned score (DLS) could replace IHC-determined PD-L1 measurement in guiding individual treatment decisions. Dall’Olio and colleagues likewise showed that total metabolic tumor volume (tMTV) ≥ 75 cm3 could mark poor prognosis in advanced NSCLC patients with high PD-L1 expression. This may help identify patients who could benefit from chemotherapy in addition to pembrolizumab.

Neoadjuvant Systemic Therapy

Neoadjuvant systemic therapy (NST) is standard care in localized and advanced breast cancer, reducing tumor size and increasing opportunities for breast-conserving surgery. However, only a few patients benefit from NST because some biologically aggressive lesions may not be controlled effectively afterward. Identifying likely beneficiaries is therefore critical. Pathological complete response (pCR) closely correlates with favorable long-term prognosis and can indicate NST effectiveness. Tsukada and colleagues predicted whether tumor types would achieve pCR after NST and identified two MRI-derived features — growth direction and contour ratio — associated with pCR. Tumors growing toward Cooper’s ligament and those with a rapid contour ratio on pretreatment multiplanar MRI predicted pCR. Kim and colleagues assessed breast cancer prognosis through recurrence-free survival, finding significantly poorer results in patients with high entropy (high heterogeneity) on T2WI.

Radiotherapy

In radiotherapy, radiogenomics may support clinical decisions through an integrated model covering both imaging and genetics. Many studies suggest it can predict radiotherapy adverse effects and identify genetic markers to help select the best regimen and maximize outcomes. Radiotherapy has also evolved from two-dimensional to three-dimensional and even four-dimensional techniques, while dose distribution has progressed from point doses to volumetric distributions. These methods depend on extensive imaging data. Radiomics provides detailed, accurate phenotyping and reveals intratumoral heterogeneity across many solid tumors. Integrating multiomics characterization into physical models also helps guide radiation physicists in expanding reirradiation protocols in targeted therapies.

Predicting Drug Response and Potential Resistance

Radiogenomics can also predict drug response and potential resistance to guide individualized tumor treatment. In 2007, Kuo and colleagues found a strong association between specific contrast-enhanced CT appearances and hepatocellular carcinoma sensitivity to adriamycin. Precision medicine requires identifying modifiable monitoring and treatment targets and developing reliable, noninvasive technologies to track changes in them over time. By providing voxel-by-voxel genetic information for a heterogeneous tumor or a group of cancers in metastatic disease, radiogenomics supports personalized treatment programs.

Radiogenomic Models as Clinical Biomarkers for Prognosis and Recurrence

Combining imaging, genetic, and pathological features into radiogenomic labels reveals links between imaging and outcomes, accelerating clinical incorporation. Developing surrogate imaging biomarkers improves clinicians’ ability to predict relevant outcomes. Radiogenomic venous invasion (RVI) is a contrast-enhanced computed tomography (CECT) biomarker of MVI. Banerjee and colleagues found that RVI on preoperative CECT was associated with poor overall survival (OS) and early recurrence. It may identify people unlikely to benefit from surgery in the long term.

Predicting Prognosis and Treatment Response with Radiogenomics

Predicting prognosis and treatment response is among radiogenomics’ most promising clinical applications. Combining tumor characteristics with genomic data provides a powerful tool for predicting outcomes and supporting decisions.

Prostate Cancer

In prostate cancer diagnosis, radiogenomics combines multiparametric MRI (mpMRI) with biopsy-based genetic data to predict progression and treatment response. Woo and colleagues found strong associations between mpMRI-derived radiogenomic data, Gleason scores, and biopsy findings, suggesting an important potential role in diagnosis and management.

Brain Tumors

In glioblastoma (GBM), radiogenomics is vital for understanding heterogeneity and developing personalized treatment plans. Models combining MRI with genomic data can predict responses and outcomes. Carrillo and colleagues developed a model predicting GBM overall survival from preoperative MR images and genomic data, with the potential to improve treatment response and long-term prognosis.

Breast Cancer

Radiogenomics is also used to predict breast cancer treatment response and develop personalized strategies. Fluid genomic data have been combined with MR features to predict neoadjuvant chemotherapy response. Sutton and colleagues showed that a model combining MR features and genomic data could predict response, highlighting radiogenomics’ importance in treatment planning.

Colorectal Cancer

Radiogenomics provides valuable information for predicting colorectal cancer (CRC) treatment response and progression. Lubner and colleagues developed a response-prediction model using genetic data and imaging features in CRC patients, with the potential to improve chemotherapy responses and long-term prognosis.

Lung Cancer

In lung cancer, radiogenomics combines genetic and imaging data to predict response and plan treatment. Radiogenomic data are used to predict responses to targeted therapies. Wei and colleagues showed that a model combining PET/CT features with genomic data could predict lung cancer treatment response and prognosis, highlighting the field’s importance.

Conclusion and Future Perspectives

Radiogenomics is emerging as an important component of personalized cancer medicine. Combining imaging and genomic data provides a powerful tool to understand heterogeneity, predict response, and optimize treatment plans. Further development and wider clinical application are expected.

Radiogenomics could revolutionize cancer research and clinical practice. Progress will contribute to more precise, effective treatments and better quality of life. It will play an important role in the future of personalized cancer medicine.

Radiogenomics: Future Directions and Challenges

Wider clinical use and acceptance require overcoming challenges involving data integration, standardization, validation, and ethics.

Data Integration

Radiogenomics generates enormous amounts of data, making integration and analysis a major challenge. Advanced computing is needed to establish links between genomic data, radiological images, and clinical information. Careful management must ensure accurate, reliable data.

Standardization

Standardizing radiogenomic data allows different research groups and clinics to compare and share information. Standardized collection, processing, and analysis will accelerate clinical integration.

Validation

Radiogenomic models need comprehensive validation before clinical use. Large, multicenter studies are required to establish reliability and accuracy. This process is important for assessing real-world performance.

Ethical Issues

Using radiogenomic data also raises privacy and ethical challenges. Patient data security and confidentiality are critical in research and practice. Data protection and ethical rules must therefore be applied rigorously.

Future Directions

Future directions in radiogenomic research will help advance personalized medicine. Some are listed below:

Artificial Intelligence and Machine Learning

Artificial intelligence (AI) and machine learning (ML) will greatly accelerate radiogenomic analysis and interpretation. Processing the resulting information faster and more accurately will support clinical decisions.

Multiomics Approaches

Combining radiogenomics with proteomic, metabolomic, and other omics data alongside genomics will make it more comprehensive and effective. Multiomics will deepen understanding of cancer biology and improve personalized strategies.

Real-Time Imaging and Treatment Planning

Radiogenomics can be used in real-time imaging and planning to monitor responses and optimize treatment at every stage. Real-time data may improve treatment effectiveness and quality of life.

Widespread Clinical Use

Radiogenomic research findings will begin to be used more widely in clinical practice, expanding personalized cancer medicine and improving treatment outcomes.

International Collaboration

International collaboration is essential to successful radiogenomic research. Sharing data and working together across countries’ research groups and clinics will accelerate progress.

Conclusion

Radiogenomics is rapidly developing into a key component of personalized cancer medicine. Integrating imaging and genomic data provides a powerful tool for understanding heterogeneity, predicting response, and optimizing treatment. It could revolutionize future cancer treatment, improve quality of life, and support more precise strategies.

Radiogenomics will continue to play an important role in personalized medicine’s future. Research and clinical applications will help develop more effective, targeted cancer treatments.