From DICOM to Dose: An Automated Python Platform for Patient-Specific CT Dosimetry.

This study introduces an automated Python-based framework for the accurate calculation of water-equivalent diameter (Dw), effective diameter (Deff), and size-specific dose estimate (SSDE) using a hybrid segmentation strategy. The algorithm integrates DICOM metadata extraction, Hounsfield Unit (HU) conversion, and interactive anatomical segmentation, combining automated HU thresholding with user-guided refinement and morphological processing to ensure robust contour delineation. The framework was validated against commercial software (Indose_CT) and expert manual segmentation using 470 clinical CT scans (270 head and 200 chest). For head scans, strong agreement was observed for Dw (r = 0.960, bias = 0.016 cm) and SSDE (r = 0.960, bias = 0.78 mGy). Chest scans demonstrated excellent accuracy, with Dw (r = 0.987, bias = 0.16 cm) and SSDE (r = 0.999, bias = 0.008 mGy). Expert validation on 100 cases confirmed near-expert segmentation performance, achieving Dice similarity coefficients of 0.966 ± 0.018 (head) and 0.953 ± 0.024 (chest), Hausdorff distances of 2.6 ± 0.8 mm (head) and 3.1 ± 1.2 mm (chest), and average surface distances below 2 mm. Importantly, segmentation discrepancies resulted in clinically negligible dosimetric differences: mean SSDE deviations were 0.360 ± 0.252 mGy (0.94%) for head and 0.142 ± 0.097 mGy (1.26%) for chest, with 98% of cases showing less than 2% deviation from expert measurements. The proposed hybrid approach effectively addresses segmentation challenges in complex anatomical cases through real-time user adjustment while maintaining efficient batch-processing capabilities via an intuitive graphical interface.