Fig. 4: Performance benchmarking comparing SpineXtract against the state-of-the-art SRH CNS classifier.

a Shows the spinal tumors test cohort composition (n = 36) used for patient-level comparison, with ependymoma being most prevalent, followed by metastasis, schwannoma, and meningioma. b Displays overall performance metrics across the entire test cohort (n = 36), demonstrating SpineXtract’s superior performance (blue bars) compared to the state-of-the-art SRH CNS classifier (gray bars) across all metrics, with particularly notable improvements in all macro-average metrics. c Provides detailed performance comparisons for each tumor type, showing SpineXtract consistently outperforming the general classifier across all four tumor categories, with particularly improvements for ependymoma sensitivity (+61.7%) and balanced accuracy (+30.8%), highlighting advantages of anatomical site-specific algorithms like SpineXtract for specialized clinical applications. d Confusion matrices demonstrate SpineXtract’s superior classification accuracy with fewer misclassifications compared to the state-of-the-art SRH CNS classifier, particularly reducing diagnostic errors and improving overall discrimination between spinal tumor types.