Journal: Frontiers in Immunology

Article Title: Non-invasive radiogenomic mapping of the SMARCAL1-driven ferroptotic niche is associated with longitudinal MRD-negative surveillance in early-stage NSCLC

doi: 10.3389/fimmu.2026.1835413

Figure Lengend Snippet: Multi-omics landscape and prognostic significance of the SMARCAL1-related ferroptotic microenvironment in baseline MRD-negative NSCLC (N = 71). (a) Region-of-interest (ROI) segmentation schematic on a representative pre-operative contrast-enhanced CT slice. ROI_core (intratumoral, solid red) was delineated by a senior thoracic radiologist (>10 years' experience) and independently verified (Dice Similarity Coefficient ≥ 0.90); ROI_rim (dashed orange) was generated by 3.0-mm morphological dilation of ROI_core with meticulous exclusion of air spaces, bones, and major vessels. Scale bar 10 mm. (b) Five-layer dimensionality reduction pipeline with embedded LASSO path: Layer 1 — 1,432 raw IBSI-compliant features filtered by intra-/inter-observer ICC ≥ 0.80 (1,105 retained); Layer 2 — coefficient of variation ≥ 0.05 and hierarchical Spearman clustering at d = 0.20 (129 retained); Layer 3 — univariate Cox P < 0.05 (42 retained); Layer 4 — LASSO Cox with nested 5×5 cross-validation (9 retained); Layer 5 — stability selection (B = 500 subsamples; Π ≥ 0.60 retained, 7/9 with Π ≥ 0.80). Right-side plots show LASSO coefficient paths and partial likelihood deviance versus log(λ). (c) Rad-Score waterfall distribution across 71 patients ranked from low to high; sustained MRD-negative patients (blue, predominantly positive values) versus MRD converters (red, predominantly negative values); maximally-selected rank cut-off at +0.15. The zero-centered distribution reflects 45 MRD-negative versus 26 MRD-converter patients. (d) Spatial sensitivity analysis of peritumoral margin width (2, 3, 5, 10 mm) showing 24-month C-index (left axis, blue circles) and Spearman ρ between Rad-Score and CIBERSORTx CD8 + T cell fraction (right axis, orange squares); optimum at 3 mm (C-index = 0.812, ρ = 0.554), supporting the a priori-selected 3-mm rim. Error bars represent bootstrap 95% CIs.

Article Snippet: Pre-operative chest CT examinations were acquired across three multi-detector CT platforms (Somatom Definition AS, Siemens Healthineers; Discovery CT750 HD, GE Healthcare; Brilliance iCT, Philips Healthcare).

Techniques: Biomarker Discovery, Generated, Selection

Journal: Frontiers in Immunology

Article Title: Non-invasive radiogenomic mapping of the SMARCAL1-driven ferroptotic niche is associated with longitudinal MRD-negative surveillance in early-stage NSCLC

doi: 10.3389/fimmu.2026.1835413

Figure Lengend Snippet: Non-invasive spatial feature extraction and five-layer construction of the Rad-Score. (a) Region-of-interest (ROI) segmentation schematic on a representative pre-operative contrast-enhanced CT slice. ROI_core (intratumoral, solid red) was delineated by a senior thoracic radiologist (>10 years’ experience) and independently verified (Dice Similarity Coefficient ≥ 0.90); ROI_rim (dashed orange) was generated by 3.0-mm morphological dilation of ROI_core with meticulous exclusion of air spaces, bones, and major vessels. Scale bar 10 mm. (b) Five-layer dimensionality reduction pipeline with embedded LASSO path: Layer 1 — 1,432 raw IBSI-compliant features filtered by intra-/inter-observer ICC ≥ 0.80 (1,105 retained); Layer 2 — coefficient of variation ≥ 0.05 and hierarchical Spearman clustering at d = 0.20 (129 retained); Layer 3 — univariate Cox P < 0.05 (42 retained); Layer 4 — LASSO Cox with nested 5×5 cross-validation (9 retained); Layer 5 — stability selection (B = 500 subsamples; Π ≥ 0.60 retained, 7/9 with Π ≥ 0.80). Right-side plots show LASSO coefficient paths and partial likelihood deviance versus log(λ). (c) Rad-Score waterfall distribution across 71 patients ranked from low to high; sustained MRD-negative patients (blue, predominantly positive values) versus MRD converters (red, predominantly negative values); maximally-selected rank cut-off at +0.15. The zero-centered distribution reflects 45 MRD-negative versus 26 MRD-converter patients. (d) Spatial sensitivity analysis of peritumoral margin width (2, 3, 5, 10 mm) showing 24-month C-index (left axis, blue circles) and Spearman ρ between Rad-Score and CIBERSORTx CD8 + T cell fraction (right axis, orange squares); optimum at 3 mm (C-index = 0.812, ρ = 0.554), supporting the a priori-selected 3-mm rim. Error bars represent bootstrap 95% CIs.

Article Snippet: Pre-operative chest CT examinations were acquired across three multi-detector CT platforms (Somatom Definition AS, Siemens Healthineers; Discovery CT750 HD, GE Healthcare; Brilliance iCT, Philips Healthcare).

Techniques: Extraction, Generated, Biomarker Discovery, Selection