Purpose To identify dosimetric parameters associated with acute hematological toxicity (HT) and identify the corresponding normal tissue complication probability (NTCP) model in cervical cancer patients receiving helical tomotherapy (Tomo) or fixed‐field intensity‐modulated radiation therapy (ff‐IMRT) in combination with chemotherapy, that is, concurrent chemoradiotherapy (CCRT) using the Lyman–Kutcher–Burman normal tissue complication probability (LKB‐NTCP) model. Methods Data were collected from 232 cervical cancer patients who received Tomo or ff‐IMRT from 2015 to 2018. The pelvic bone marrow (PBM) (including the ilium, pubes, ischia, acetabula, proximal femora, and lumbosacral spine) was contoured from the superior boundary (usually the lumbar 5 vertebra) of the planning target volume (PTV) to the proximal end of the femoral head (the lower edge of the ischial tubercle). The parameters of the LKB model predicting ≥grade 2 hematological toxicity (Radiation Therapy Oncology Group [RTOG] grading criteria) (TD50(1), m, and n) were determined using maximum likelihood analyses. Univariate and multivariate logistic regression analyses were used to identify correlations between dose–volume parameters and the clinical factors of HT. Results In total, 212 (91.37%) patients experienced ≥grade 2 hematological toxicity. The fitted normal tissue complication probability model parameters were TD50(1) = 38.90 Gy (95%CI, [36.94, 40.96]), m = 0.13 (95%CI [0.12, 0.16]), and n = 0.04 (95%CI [0.02, 0.05]). Per the univariate analysis, the NTCP (the use of LKB‐NTCP with the set of model parameters found, p = 0.023), maximal PBM dose (p = 0.01), mean PBM dose (p = 0.021), radiation dose (p = 0.001), and V16–53 (p < 0. 05) were associated with ≥grade 2 HT. The NTCP (the use of LKB‐NTCP with the set of model parameters found, p = 0.023; AUC = 0.87), V16, V17, and V18 ≥ 79.65%, 75.68%, and 72.65%, respectively (p < 0.01, AUC = 0.66∼0.68), V35 and V36 ≥ 30.35% and 28.56%, respectively (p < 0.05; AUC = 0.71), and V47 ≥ 13.43% (p = 0.045; AUC = 0.80) were significant predictors of ≥grade 2 hematological toxicity from the multivariate logistic regression analysis. Conclusions The volume of the PBM of patients treated with concurrent chemoradiotherapy and subjected to both low‐dose (V16–18) and high‐dose (V35,36 and V47) irradiation was associated with hematological toxicity, depending on the fractional volumes receiving the variable degree of dosage. The NTCP were stronger predictors of toxicity than V16–18, V35, 36, and V47. Hence, avoiding radiation hot spots on the PBM could reduce the incidence of severe HT.

Background Routinely delineating of important skeletal growth centers is imperative to mitigate radiation‐induced growth abnormalities for pediatric cancer patients treated with radiotherapy. However, it is hindered by several practical problems, including difficult identification, time consumption, and inter‐practitioner variability. Purpose The goal of this study was to construct and evaluate a novel Triplet‐Attention U‐Net (TAU‐Net)‐based auto‐segmentation model for important skeletal growth centers in childhood cancer radiotherapy, concentrating on the accuracy and time efficiency. Methods A total of 107 childhood cancer patients fulfilled the eligibility criteria were enrolled in the training cohort (N = 80) and test cohort (N = 27). The craniofacial growth plates, shoulder growth centers, and pelvic ossification centers, with a total of 19 structures in the three groups, were manually delineated by two experienced radiation oncologists on axial, coronal, and sagittal computed tomography images. Modified from U‐Net, the proposed TAU‐Net has one main branch and two bypass branches, receiving semantic information of three adjacent slices to predict the target structure. With supervised deep learning, the skeletal growth centers contouring of each group was generated by three different auto‐segmentation models: U‐Net, V‐Net, and the proposed TAU‐Net. Dice similarity coefficient (DSC) and Hausdorff distance 95% (HD95) were used to evaluate the accuracy of three auto‐segmentation models. The time spent on performing manual tasks and manually correcting auto‐contouring generated by TAU‐Net was recorded. The paired t‐test was used to compare the statistical differences in delineation quality and time efficiency. Results Among the three groups, including craniofacial growth plates, shoulder growth centers, and pelvic ossification centers groups, TAU‐Net had demonstrated highly acceptable performance (the average DSC = 0.77, 0.87, and 0.83 for each group; the average HD95 = 2.28, 2.07, and 2.86 mm for each group). In the overall evaluation of 19 regions of interest (ROIs) in the test cohort, TAU‐Net had an overwhelming advantage over U‐Net (63.2% ROIs in DSC and 31.6% ROIs in HD95, p = 0.001–0.042) and V‐Net (94.7% ROIs in DSC and 36.8% ROIs in HD95, p = 0.001–0.040). With an average time of 52.2 min for manual delineation, the average time saved to adjust TAU‐Net‐generated contours was 37.6 min (p < 0.001), a 72% reduction. Conclusions Deep learning–based models have presented enormous potential for the auto‐segmentation of important growth centers in pediatric skeleton, where the proposed TAU‐Net outperformed the U‐Net and V‐Net in geometrical precision for the majority status.