Enhancing Spinal Surgery Through Pyradiomics: A Systematic Review of Its Use, Understandability, and Future Directions

Introduction: Computer Vision (CV) is a branch of artificial intelligence that allows computers to interpret and analyze visual data, mimicking human visual perception to automate complex tasks. Within the realm of medical imaging, radiomics emerges as a crucial application of CV, extracting quantifiable features from images to reflect underlying biological properties. However, radiomics struggles with standardization issues, affecting the consistency and comparability of studies. This study focuses on PyRadiomics, an open-source Python platform developed to overcome these challenges by providing robust feature extraction capabilities, and its application for spinal surgery research. We review how PyRadiomics enhances clinical decision-making and discuss its prospective roles in the evolution of spine surgery.

Methods: Literature review was conducted using PubMed, Web of Science, and Embase, following PRISMA guidelines. Search terms included combinations of spine, surgery, and radiomics. Articles were screened for duplicates, and title and abstract reviews performed. Full-text articles meeting inclusion criteria were analyzed.

Results: From 139 articles, 10 retrospective studies were included, with sample sizes ranging from 62 to 386 (average: 160). SVM was the most frequently used classifier (50%), followed by Random Forest and Logistic Regression. Wavelet and GLCM filters were commonly applied, with all studies utilizing PyRadiomics’ standard features, such as shape and texture. None of the studies addressed model interpretability.
Performance metrics were strong, with AUCs ranging from 0.78 to 0.96 (average: 0.85) and accuracy exceeding 90% in several models. PyRadiomics improved diagnosis, prognosis, and treatment planning for spinal conditions, including predicting tumor aggressiveness, assessing bone health, and stratifying surgical risk. These applications contributed to personalized, effective surgical interventions and reduced the need for invasive procedures.

Conclusion : PyRadiomics shows promise in spinal surgery diagnostic and predictive modeling, enhancing diagnostic accuracy and enabling personalized treatment planning. Standardizing feature extraction, improving AI explainability, and prospective validation are crucial next steps for its broader clinical adoption.