Delta-Radiomics

The core mechanism involves computing the mathematical difference (Δ) between radiomic features extracted from a baseline scan and a follow-up scan. This process requires rigid and deformable image registration to ensure voxel-wise correspondence across time points.

• Formula: ΔFeature = Feature(Time_1) - Feature(Time_0)
• Registration: Corrects for patient positioning and anatomical changes between scans.
• Harmonization: ComBat harmonization is critical to remove scanner-specific technical drift before calculating deltas.

Delta-radiomic features serve as early surrogate endpoints for pathological complete response (pCR) and overall survival. Unlike static tumor measurements (RECIST), delta features capture microstructural changes in tumor heterogeneity that precede macroscopic shrinkage.

• Early Assessment: Predicts treatment efficacy within days or weeks of initiation.
• Heterogeneity Capture: A decrease in Entropy and Cluster Prominence often indicates a positive response to chemotherapy.
• Immunotherapy: Delta-radiomics can differentiate true progression from pseudoprogression in checkpoint inhibitor therapy.

Texture matrices (GLCM, GLRLM, GLSZM) are particularly sensitive to longitudinal changes. A reduction in Short Run Emphasis (SRE) or Gray-Level Non-Uniformity (GLN) often correlates with the homogenization of tumor tissue during successful treatment.

• GLCM Changes: Delta-Contrast and Delta-Correlation reflect alterations in local intensity variations.
• GLSZM Changes: A decrease in Zone Percentage (ZP) indicates the fragmentation of necrotic regions.
• Wavelet Decomposition: Delta features are often calculated on wavelet-filtered sub-bands to isolate changes at specific spatial frequencies.

Longitudinal studies are highly susceptible to the 'batch effect'—non-biological variance introduced by scanner software upgrades or reconstruction algorithm changes. ComBat harmonization is applied to feature distributions before subtraction to isolate true biological change.

• Scanner Drift: Mitigates variance from hardware calibration shifts over months.
• Protocol Standardization: Essential when baseline and follow-up scans are acquired on different vendors (GE, Siemens, Philips).
• IBSI Compliance: Harmonization pipelines must adhere to the Image Biomarker Standardisation Initiative to ensure delta reproducibility.

Not all radiomic features are suitable for delta analysis. Features with low Intraclass Correlation Coefficient (ICC) test-retest reliability are excluded. LASSO and mRMR are used to select the most stable and predictive delta features.

• Stability Filtering: Only features with ICC > 0.8 on test-retest scans are considered.
• Delta-Radiomic Signature: A composite biomarker combining multiple delta features (e.g., ΔSphericity + ΔEntropy) to predict progression-free survival.
• Overfitting Prevention: Nested cross-validation is mandatory due to the high dimensionality of delta feature sets.

Delta-radiomics extends to habitat imaging, where tumors are partitioned into distinct sub-regions (e.g., enhancing, necrotic, edematous) based on voxel-wise clustering. Tracking the volume and textural shifts of these habitats provides a high-resolution map of therapeutic resistance.

• Habitat Shrinkage: Quantifies the differential response of tumor sub-regions.
• Virtual Biopsy: Non-invasively tracks the evolution of the tumor microenvironment.
• Spatial Delta Maps: Visualizes exactly where within the tumor the most significant textural changes are occurring.