Background <p>Standardized lesion segmentation on ⁶⁸Ga-Pentixafor PET-CT is lacking in multiple myeloma (MM), and it is unclear how different segmentation methods affect semi-quantitative PET features. This study aimed to evaluate semi-automatic segmentation strategies and their impact on total lesion expression of C-X-C chemokine receptor type 4 (TLE_CXCR4) quantification and prognostic value.</p> Methods <p>We retrospectively analyzed ⁶⁸Ga-Pentixafor PET-CT scans from 49 newly diagnosed MM patients. Skeletal volume of interest was automatically delineated using an AI-based tool, and six semi-automatic segmentation approaches were applied using SUV thresholds based on liver, spleen, and bone uptake. Quality scores for lesion coverage, non-target inclusion, and contour fit were assessed. TLE_CXCR4 and other parameters were compared, and their associations with bone marrow plasma cell infiltration and 3-year disease progression were analyzed.</p> Results <p>Segmentation performance varied widely. The liver2× approach achieved the best balance between lesion coverage and non-target marking. TLE_CXCR4 from all approaches correlated moderately with bone marrow infiltration (ρ = 0.31–0.54), and strongly with visual staging (ρ = 0.77–0.88, except for bone40%). The liver2×-derived TLE_CXCR4 yielded the highest predictive accuracy for disease progression (AUC = 0.811), outperforming traditional total-bone segmentation, though not significantly.</p> Conclusion <p>Among various segmentation strategies, liver2× offers optimal performance for semi-automatic lesion delineation on ⁶⁸Ga-Pentixafor PET-CT in MM. TLE_CXCR4 derived from this method correlates well with tumor burden and prognostic outcomes, supporting its clinical applicability for risk stratification.</p> Key points <p>Semi-automatic segmentation strategies on ⁶⁸Ga-Pentixafor PET-CT were systematically evaluated in newly diagnosed multiple myeloma patients, whole skeletal delineation and SUV-based thresholds from liver, spleen, and bone.</p> <p>Among six methods, the liver2× approach provided the best trade-off between lesion coverage and avoidance of non-target uptake and generated total lesion expression of C-X-C chemokine receptor type 4 (TLE_CXCR4) values strongly correlated with bone marrow plasma cell infiltration (ρ up to 0.54) and PET-based visual staging (ρ up to 0.88).</p> <p>Twice the liver SUVmean (liver2×)-derived TLE_CXCR4achieved the highest predictive accuracy for 3-year progression (AUC = 0.811), supporting its clinical utility as a standardized, semi-automatic approach for risk stratification in multiple myeloma.</p>