Radiomics meets pathology: Building the bridge for precision oncology

Dear Editor,

Modern oncology faces a crucial challenge: delivering accurate diagnoses that capture the dynamic and heterogeneous biology of cancer. Tissue biopsy remains the reference standard for detecting driver mutations, chromosomal rearrangements, and epigenetic alterations. However, it is invasive, not always repeatable, and inherently limited by tumor spatial and temporal heterogeneity.^[1,2]

Radiomics has emerged as a complementary strategy, transforming routine medical images into quantitative biomarkers. When integrated with molecular and histopathological data, radiomics enables tumor phenotype prediction and aligns imaging with the principles of precision medicine.^[1-3] Radiogenomics, linking radiomic features with genomic data, has further expanded this potential, with applications across non-small-cell lung cancer (NSCLC), gliomas, breast cancer, gastrointestinal malignancies, and head-and-neck tumors.^[4-8]

A prospective study by Felfli et al. exemplifies this paradigm. They extracted radiomic features from pre-operative computed tomography (CT) scans of NSCLC patients. They compared them with mutation profiles derived from circulating tumor DNA, a minimally invasive tool that captures tumor evolution in real time.^[5] Second-order texture features proved predictive for single mutations, although performance declined with more complex mutational states, underscoring both biological complexity and the need for more sophisticated algorithms.^[8]

In NSCLC, epidermal growth factor receptor (EGFR) mutation prediction has been the most robust application (area under the curve 0.75-0.94), while kirsten rat sarcoma (KRAS) and anaplastic lymphoma kinase (ALK) remain more challenging.^[6] Multiparametric magnetic resonance imaging has shown promise in predicting EGFR status in brain metastases, extending radiomics’ utility into metastatic disease.^[7,8]

Importantly, the bridge between radiology and pathology is not merely conceptual. Ottaiano et al. demonstrated that radiomic analysis of CT-guided fine-needle aspiration cytology in lung cancer could correlate cytomorphological patterns (solid, papillary, and mixed) with imaging-derived metrics such as kurtosis and skewness.^[7] Such integration can improve diagnostic yield, guide biopsy targeting, and reduce unnecessary invasive procedures.

Beyond thoracic oncology, radiomics has predicted isocitrate dehydrogenase (IDH) mutation and MGMT methylation in gliomas, receptor status in breast cancer, and treatment response in gastrointestinal malignancies. Multicenter initiatives, including those using The Cancer Imaging Archive, have confirmed that harmonized pipelines can achieve reproducible results across institutions.^[9]

However, clinical translation remains a major hurdle. Several studies are retrospective and single-center, and variability in acquisition protocols undermines generalizability. Moreover, machine learning models often operate as “black boxes,” limiting clinical trust. Progress will depend on standardized workflows (e.g., Image Biomarker Standardisation Initiative (IBSI) compliance), harmonization methods (e.g., ComBat), and explainable Artificial Intelligenceto illuminate the biological rationale behind predictions.^[10]

The future of radiomics lies in prospective clinical trials, particularly in targeted therapy contexts where molecular status dictates treatment. Here, radiomics could serve as both a non-invasive screening tool and a dynamic monitor of tumor evolution. Integration with digital pathology and multi-omic platforms promises composite biomarkers that surpass the predictive power of any single modality.