Brian S Tao, Katelyn A Tao, Mario Keko, Nazim Haouchine, Ron Noah Alkalay
Machine learning models based on CT-derived radiomic features may enable accurate and noninvasive early prediction of treatment response in patients with advanced NSCLC receiving chemo-immunotherapy or immunotherapy alone, supporting the potential role of radiomics as an imaging biomarker for treatment stratification and personalized therapeutic planning.
Radiomics has emerged as a transformative tool in medical imaging, offering the ability to extract complex quantitative features that are generally indiscernible to the human eye. These radiomic features can be interrogated non-invasively on medical imaging. Importantly, in contrast to traditional diagnostic methods reliant on a single, often scalar, measure, radiomics features can form a high-dimensional data space from imaging data suitable for machine learning. Within the framework of artificial intelligence, radiomic features can be harnessed for differentiating healthy from pathological tissue, risk-stratifying patients for benign and malignant fractures, and clinical outcome measures. This review presents an introduction to the methodology underlying radiomics feature selection, reproducibility, feature analysis and model building, assessment of model performance, and open-source libraries for extracting radiomics features from imaging. The review then highlights the application of radiomics in osseous and cartilaginous spinal imaging for identifying osteoporosis and the prediction of fragility vertebral fractures, chronic low back pain and the assessment of intervertebral disc degeneration and herniation, cancer metastatic spine disease and the differentiation of benign vs. malignant lesions and the classification of benign vs. malignant vertebral fractures. Throughout this stage, we endeavor to demonstrate how radiomics can analyze imaging biomarkers to detect subtle structural changes in vertebral bone microarchitecture, assess tissue quality and early-stage fractures, and identify radiomic biomarkers of low back pain chronicity and intervertebral disc heterogeneity that signal degeneration and herniation risk. The review culminates with the presentation of the current limitations and future research directions, including opportunities for integration with multi-omics analysis, highlighting radiomics' potential for enhanced diagnostic accuracy and more personalized patient care. The application of radiomics in spinal imaging offers a promising avenue for improving non-invasive image early detection, risk stratification, and personalized management of spinal pathology, paving the way for more effective interventions and improved patient outcomes.