Nasopharyngeal cancer adaptive radiotherapy with CBCT-derived synthetic CT: deep learning-based auto-segmentation precision and dose calculation consistency on a C-Arm linac

Abstract

Background: To evaluate the precision of automated segmentation facilitated by deep learning (DL) and dose calculation in adaptive radiotherapy (ART) for nasopharyngeal cancer (NPC), leveraging synthetic CT (sCT) images derived from cone-beam CT (CBCT) scans on a conventional C-arm linac.

Materials and methods: Sixteen NPC patients undergoing a two-phase offline ART were analyzed retrospectively. The initial (pCT₁) and adaptive (pCT₂) CT scans served as gold standard alongside weekly acquired CBCT scans. Patient data, including manually delineated contours and dose information, were imported into ArcherQA. Using a cycle-consistent generative adversarial network (cycle-GAN) trained on an independent dataset, sCT images (sCT₁, sCT₄, sCT₄^*) were generated from weekly CBCT scans (CBCT₁, CBCT₄, CBCT₄) paired with corresponding planning CTs (pCT₁, pCT₁, pCT₂). Auto-segmentation was performed on sCTs, followed by GPU-accelerated Monte Carlo dose recalculation. Auto-segmentation accuracy was assessed via Dice similarity coefficient (DSC) and 95th percentile Hausdorff distance (HD₉₅). Dose calculation fidelity on sCTs was evaluated using dose-volume parameters. Dosimetric consistency between recalculated sCT and pCT plans was analyzed via Spearman's correlation, while volumetric changes were concurrently evaluated to quantify anatomical variations.

Results: Most anatomical structures demonstrated high pCT-sCT agreement, with mean values of DSC > 0.85 and HD₉₅ < 5.10 mm. Notable exceptions included the primary Gross Tumor Volume (GTVp) in the pCT₂-sCT₄ comparison (DSC: 0.75, HD₉₅: 6.03 mm), involved lymph node (GTVn) showing lower agreement (DSC: 0.43, HD₉₅: 16.42 mm), and submandibular glands with moderate agreement (DSC: 0.64-0.73, HD₉₅: 4.45-5.66 mm). Dosimetric analysis revealed the largest mean differences in GTVn D₉₉: -1.44 Gy (95% CI: [-3.01, 0.13] Gy) and right parotid mean dose: -1.94 Gy (95% CI: [-3.33, -0.55] Gy, p < 0.05). Anatomical variations, quantified via sCTs measurements, correlated significantly with offline adaptive plan adjustments in ART. This correlation was strong for parotid glands (ρ > 0.72, p < 0.001), a result that aligned with sCT-derived dose discrepancy analysis (ρ > 0.57, p < 0.05).

Conclusion: The proposed method exhibited minor variations in volumetric and dosimetric parameters compared to prior treatment data, suggesting potential efficiency improvements for ART in NPC through reduced human dependency.

Keywords: Adaptive radiotherapy; Automated segmentation; Deep learning; Nasopharyngeal cancer; Synthetic CT.

Fig. 1

Box plots of relative volumetric variations across different image sets. The numbers on the x-axis represent the group indices. The three groups were the following, each referenced to pCT₁: group 1=(pCT₂-pCT₁)/pCT₁, group 2= (sCT₄-pCT₁)/pCT₁, group 3= (sCT₄*-pCT₁)/pCT₁. Abbreviations: PG = the parotid glands; OC = the oral cavity; SG = the submandibular glands. The values displayed above the boxes represent the calculated Spearman correlation coefficients. The mean values and 95%CI displayed beneath the sub-figures were statistically analyzed using paired two-tailed t-test. Asterisks denote statistical significance (p < 0.05). The box spans the interquartile range (IQR; 25th−75th percentiles). Whiskers extend to the 1st and 99th percentiles, with individual circles representing outliers beyond this range. The line and rectangle represent the median and mean values, respectively

Fig. 2

Box plots of the dosimetric difference of seven distinct combinations of plans and scans. The seven distinct combinations of plans and scans were as follows. Ori-plan: plan1 on pCT₁, ori-test-plan: plan1 on sCT₁; adap-plan: plan2 on pCT₂, adap-test-plan: plan2 on sCT₄^*; recal-plan: plan1 on pCT₂, recal-test-plan1: plan1 on sCT₄, recal-test-plan2: plan1 on sCT₄^*. The numbers on the x-axis represent the group indices. The group 1 and 2 represent the dosimetric accuracy evaluated relative to the offline ART treatment plans (group 1: ori-test-plan vs. ori-plan, group 2: adap-test-plan vs. adap-plan). The group 3 represents the dose deviation of recal-plan relative to ori-plan. The group 4 illustrates the improvement of the adaptive plan (adap-plan) over recal-plan. The group 5 (recal-test-plan1 vs. ori-plan) depicts the dosimetric agreement with group 3 (recal-plan vs. ori-plan), which reflects the ART evaluation performance. The group 6 (recal-test-plan2 vs. ori-plan) represents a verification of group 5. Abbreviations: MD = mean dose, PG = the parotid glands, OC = the oral cavity, SG = the submandibular glands. The values displayed above the boxes represent the calculated Spearman correlation coefficients. The mean values and 95%CI displayed beneath the sub-figures were statistically analyzed using paired two-tailed t-test. Asterisks denote statistical significance (p < 0.05). The box spans the interquartile range (IQR; 25th–75th percentiles). Whiskers extend to the 1st and 99th percentiles, with individual circles representing outliers beyond this range. The line and rectangle represent the median and mean values, respectively

Fig. 3

One case of structure delineations, dose distributions and DVHs using the pCTs and sCTs. (a) pCT₁ and dose distribution of the ori-plan, (b) pCT₂ and dose distribution of the recal-plan, (c) sCT₁ and dose distribution of the ori-test-plan, (d) sCT₄ and dose distribution of the recal-test-plan, (e) DVH comparison between the ori-plan and the ori-test-plan, (f) DVH comparison between the recal-plan and the recal-test-plan1