1250 - Mid-Treatment CT Mean Density, Volume and Mass Changes as Early Response Biomarkers in Head and Neck Cancer Radiotherapy Using AI-Based Deformable Image Registration

Purpose/Objective(s):

Early, quantitative assessment of tumor regression during radiotherapy (RT) is essential for adaptive treatment planning and digital-twin model development in head and neck cancer (HNC). This study characterizes mid-treatment changes in CT-derived tumor density, volume, and mass using an AI-based deformable image registration (DIR) framework (ProRSeg) and evaluates Jacobian-derived deformation metrics as early imaging biomarkers of regression with the long-term goal of predicting local control outcomes.

Materials/Methods:

Paired pre-treatment (T0) and mid-treatment (T1, ~3 weeks into RT) CT scans from 104 HNC patients were analyzed. Following rigid alignment, DIR was performed using ANTs and ProRSeg (a 3D convolutional LSTM with U-Net encoder fine-tuned on 62 patients) and validated via Dice similarity coefficient (DSC) and 95th-percentile Hausdorff distance (HD95). Mean CT density, volume, and mass were computed within primary and lymph node at T0 and T1. Percent change per day was derived for all three metrics. Jacobian histogram statistics were correlated with response metrics using Spearman correlation. Wilcoxon signed-rank tests assessed T0-to-T1 changes.

Results:

ProRSeg outperformed ANTs (DSC: 0.74 vs. 0.62; HD95: 5.0 vs. 6.5 mm). Significant regression was observed by ~3 weeks across all metrics in both regions (Table 1; all p<0.001). Response was heterogeneous, about 67–77% of patients showed reductions in volume, mass, and CT density. Lymph nodes showed greater density loss than primary tumors (-0.033 vs. -0.016 %/day), while volume (-0.46 vs. -0.44) and mass loss (-0.50 vs. -0.55 %/day) were comparable. Jacobian mean correlated with lymph node volume and mass loss (?~-0.4, p<0.001) but not primary. Jacobian SD correlated with density change in primary (?~-0.30, p<0.05) and lymph nodes (?~-0.52, p<0.001).

Table 1. Tumor Regression Metrics at Mid-Treatment

†Wilcoxon signed-rank test for T0?T1 change; ?^(Spearman) = Spearman correlation coefficient

Conclusion:

Significant tumor regression in CT mean density, volume, and mass is detectable by ~3 weeks into RT in both primary HNC tumors and lymph nodes, with significant variability between patients. These early tumor changes should be further tested as biomarkers for predicting local control and developing digital-twin based adaptive radiotherapy strategies.

Metrics	Primary Tumor mean ± SD [median]	Lymph Nodes mean ± SD [median]	p-value (T0?T1)† Primary [Lymph Nodes]
% ??mean/day	-0.016 ± 0.034 [-0.011]	-0.033 ± 0.047 [-0.027]	<0.001 [<0.001]
% ?V/day	-0.44 ± 0.62 [-0.44]	-0.46 ± 0.49 [-0.45]	<0.001 [<0.001]
%?M/day	-0.55 ± 0.72 [-0.55]	-0.50 ± 0.51 [-0.49]	<0.001 [<0.001]
?(Spearman) (|J|mean and % ?V/day)	-0.09 (p-value > 0.05)	-0.41(p-value < 0.001)	—
?(Spearman) (|J|mean and % ?M/day)	-0.08 (p-value > 0.05)	-0.40 (p-value < 0.001)	—
?(Spearman) (|J|var and % ??mean/day)	-0.18 (p-value > 0.05)	-0.52 (p-value < 0.001)	—
?(Spearman) (|J|SD and % ??mean/day)	-0.30 (p-value < 0.05)	-0.52 (p-value < 0.001)	—