Differentiating Tuberculous and Pyogenic Spondylodiscitis: Part II: MRI/CT Features, Prediction Models, and the Role of PET/CT—A Narrative Review

Distinguishing tuberculous spondylodiscitis (TS) from pyogenic spondylodiscitis (PS) on imaging remains clinically important because the two entities differ in antimicrobial strategy, surgical timing, and the urgency of microbiological workup, yet no single imaging sign is pathognomonic for either diagnosis. When tissue confirmation is delayed, imaging becomes the primary tool that shifts pre-test probability and guides biopsy strategy—a role that is particularly critical in culture-negative disease, in antibiotic-exposed patients, and in settings with limited access to rapid mycobacterial diagnostics. This narrative review maps the imaging evidence most useful for routine TS-versus-PS differentiation, synthesizes recent prediction models and quantitative tools, and evaluates the adjunctive role of 18F-FDG PET/CT in biopsy-oriented decision-making. PubMed/MEDLINE was searched from inception to 31 March 2026 using pre-specified search terms across three concept blocks. This was a narrative review without formal pooling of quantitative estimates or prospective protocol registration. A secondary Scopus search identified no additional eligible records. Priority was given to comparative TS-versus-PS imaging cohorts, meta-analyses of MRI and CT features, prediction-model studies, and the recent radiomics and PET/CT literature; radiomics and deep-learning studies are appraised as a distinct evidence tier given their retrospective single-center derivation and predominantly internal-only validation. In total, 14 comparative TS-versus-PS imaging cohorts, three meta-analyses, and 14 prediction-model, quantitative-scoring, radiomics, or deep-learning studies formed the core evidence base. MRI yields the greatest differentiating information and is the recommended first-line modality. Features favoring TS include thoracic predominance, multilevel or non-contiguous involvement, relative early disc preservation, subligamentous spread, intraosseous abscesses, thin-walled paravertebral collections, and severe vertebral collapse with kyphotic deformity. Features favoring PS include lumbar predominance, early disc-endplate destruction, homogeneous inflammatory enhancement, facet-joint arthritis, and epidural phlegmon. CT adds osseous detail—including sequestra, subligamentous bone erosion, and paravertebral calcification—and remains the primary guidance modality for biopsy. Recent prediction models combining imaging and laboratory variables have shown high discriminative performance in derivation cohorts; however, most remain internally validated and locally calibrated. Pending external validation, none should be considered ready for unmodified clinical adoption, and they should therefore be used as probability modifiers rather than stand-alone diagnostic rules. 18F-FDG PET/CT is complementary rather than primary and is most useful when MRI is contraindicated or equivocal, in hardware-associated infection, and for whole-body staging in suspected disseminated tuberculosis. Non-infectious mimics and alternative infectious diagnoses should be considered when the imaging pattern is internally inconsistent or when standard cultures remain negative. Imaging findings should refine etiological probability and guide biopsy strategy, but should not replace tissue confirmation when tissue is feasible. Non-imaging evidence complementing this review is addressed in the companion manuscript, Part I.