Department of Radiation Medicine and Applied Sciences University of California San Diego La Jolla, CA.
Department of Radiation Medicine and Applied Sciences University of California San Diego La Jolla, CA.
Brachytherapy. 2022 Jul-Aug;21(4):532-542. doi: 10.1016/j.brachy.2022.03.002. Epub 2022 May 11.
The purpose of this work was to develop a knowledge-based dose prediction system using a convolution neural network (CNN) for cervical brachytherapy treatments with a tandem-and-ovoid applicator.
A 3D U-NET CNN was utilized to make voxel-wise dose predictions based on organ-at-risk (OAR), high-risk clinical target volume (HRCTV), and possible source location geometry. The model comprised 395 previously treated cases: training (273), validation (61), test (61). To assess voxel prediction accuracy, we evaluated dose differences in all cohorts across the dose range of 20-130% of prescription, mean (SD) and standard deviation (σ), as well as isodose dice similarity coefficients for clinical and/or predicted dose distributions. We examined discrete Dose-Volume Histogram (DVH) metrics utilized for brachytherapy plan quality assessment (HRCTV D90%; bladder, rectum, and sigmoid D2cc) with ΔD=D-D mean, standard deviation, and Pearson correlation coefficient further quantifying model performance.
Ranges of voxel-wise dose difference accuracy (δD¯±σ) for 20-130% dose interval in training (test) sets ranged from [-0.5% ± 2.0% to +2.0% ± 14.0%] ([-0.1% ± 4.0% to +4.0% ± 26.0%]) in all voxels, [-1.7% ± 5.1% to -3.5% ± 12.8%] ([-2.9% ± 4.8% to -2.6% ± 18.9%]) in HRCTV, [-0.02% ± 2.40% to +3.2% ± 12.0%] ([-2.5% ± 3.6% to +0.8% ± 12.7%]) in bladder, [-0.7% ± 2.4% to +15.5% ± 11.0%] ([-0.9% ± 3.2% to +27.8% ± 11.6%]) in rectum, and [-0.7% ± 2.3% to +10.7% ± 15.0%] ([-0.4% ± 3.0% to +18.4% ± 11.4%]) in sigmoid. Isodose dice similarity coefficients ranged from [0.96,0.91] for training and [0.94,0.87] for test cohorts. Relative DVH metric prediction in the training (test) set were HRCTV ΔD¯±σ = -0.19 ± 0.55Gy (-0.09 ± 0.67 Gy), bladder ΔD¯±σ = -0.06 ± 0.54Gy (-0.17 ± 0.67 Gy), rectum ΔD¯±σ= -0.03 ± 0.36Gy (-0.04 ± 0.46 Gy), and sigmoid ΔD¯±σ = -0.01 ± 0.34Gy (0.00 ± 0.44 Gy).
A 3D knowledge-based dose predictions provide voxel-level and DVH metric estimates that could be used for treatment plan quality control and data-driven plan guidance.
本研究旨在开发一种基于卷积神经网络（CNN）的宫颈癌近距离治疗中使用 tandem-and-ovoid 施源器的剂量预测系统。
使用 3D U-NET CNN 进行体素剂量预测，基于危及器官（OAR）、高危临床靶区（HRCTV）和可能的源位置几何形状。该模型包含 395 例既往治疗病例：训练（273 例）、验证（61 例）和测试（61 例）。为了评估体素预测的准确性，我们评估了所有队列在 20-130%处方剂量范围内的剂量差异，平均值（SD）和标准差（σ），以及临床和/或预测剂量分布的等剂量骰子相似系数。我们检查了用于近距离治疗计划质量评估的离散剂量-体积直方图（DVH）指标（HRCTV D90%；膀胱、直肠和乙状结肠 D2cc），使用ΔD=D-D mean、标准差和 Pearson 相关系数进一步量化了模型性能。
在训练（测试）组 20-130%剂量间隔的体素剂量差异准确性（δD¯±σ）范围为[-0.5%±2.0%至+2.0%±14.0%]（[-0.1%±4.0%至+4.0%±26.0%]）在所有体素中，HRCTV 中为[-1.7%±5.1%至-3.5%±12.8%]（[-2.9%±4.8%至-2.6%±18.9%]），膀胱中为[-0.02%±2.40%至+3.2%±12.0%]（[-2.5%±3.6%至+0.8%±12.7%]），直肠中为[-0.7%±2.4%至+15.5%±11.0%]（[-0.9%±3.2%至+27.8%±11.6%]），乙状结肠中为[-0.7%±2.3%至+10.7%±15.0%]（[-0.4%±3.0%至+18.4%±11.4%]）。等剂量骰子相似系数范围为训练组[0.96，0.91]和测试组[0.94，0.87]。在训练（测试）组中，相对 DVH 指标预测为 HRCTV ΔD¯±σ=-0.19±0.55Gy（-0.09±0.67 Gy），膀胱 ΔD¯±σ=-0.06±0.54Gy（-0.17±0.67 Gy），直肠 ΔD¯±σ=-0.03±0.36Gy（-0.04±0.46 Gy），乙状结肠 ΔD¯±σ=-0.01±0.34Gy（0.00±0.44 Gy）。
基于知识的三维剂量预测可以提供体素水平和 DVH 指标估计值，可用于治疗计划质量控制和数据驱动的计划指导。
