Purpose

Automated tools can help identify radiation treatment plans of unacceptable quality. To this end, we developed a quality verification technique to automatically verify the clinical acceptability of beam apertures for 4-field box treatments of patients with cervical cancer. By comparing the beam apertures to be used for treatment with a secondary set of beam apertures developed automatically, this quality verification technique can flag beam apertures that may need to be edited to be acceptable for treatment.

Methods and Materials

The automated methodology for creating verification beam apertures uses a deep learning model trained on beam apertures and digitally reconstructed radiographs from 255 clinically acceptable planned treatments (as rated by physicians). These verification apertures were then compared with the treatment apertures using spatial comparison metrics to detect unacceptable treatment apertures. We tested the quality verification technique on beam apertures from 80 treatment plans. Each plan was rated by physicians, where 57 were rated clinically acceptable and 23 were rated clinically unacceptable.

Results

Using various comparison metrics (the mean surface distance, Hausdorff distance, and Dice similarity coefficient) for the 2 sets of beam apertures, we found that treatment beam apertures rated acceptable had significantly better agreement with the verification beam apertures than those rated unacceptable (P < .01). Upon receiver operating characteristic analysis, we found the area under the curve for all metrics to be 0.89 to 0.95, which demonstrated the high sensitivity and specificity of our quality verification technique.

Conclusions

We found that our technique of automatically verifying the beam aperture is an effective tool for flagging potentially unacceptable beam apertures during the treatment plan review process. Accordingly, we will clinically deploy this quality verification technique as part of a fully automated treatment planning tool and automated plan quality assurance program.

中文翻译：

---

自动验证宫颈癌放射治疗的射束孔径。

目的

自动化工具可以帮助识别质量不可接受的放射治疗计划。为此，我们开发了一种质量验证技术，可以自动验证光束孔径对宫颈癌患者进行 4 视场盒治疗的临床可接受性。通过将用于治疗的射束孔径与自动开发的第二组射束孔径进行比较，这种质量验证技术可以标记可能需要编辑以适合治疗的射束孔径。

方法和材料

创建验证射束孔径的自动化方法使用了深度学习模型，该模型在射束孔径和来自 255 种临床可接受的计划治疗（由医生评分）的数字重建放射线照片上进行训练。然后使用空间比较度量将这些验证孔径与治疗孔径进行比较，以检测不可接受的治疗孔径。我们测试了 80 个治疗计划的射束孔径的质量验证技术。每个计划均由医生进行评级，其中 57 项被评为临床可接受，23 项被评为临床不可接受。

结果

使用 2 组射束孔径的各种比较指标（平均表面距离、Hausdorff 距离和 Dice 相似系数），我们发现，与那些被评为不可接受的治疗射束孔径相比，被评为可接受的治疗射束孔径与验证射束孔径的一致性明显更好（P < .01）。在接受者操作特征分析中，我们发现所有指标的曲线下面积为 0.89 至 0.95，这证明了我们的质量验证技术的高灵敏度和特异性。

结论

我们发现，我们的自动验证射束孔径的技术是在治疗计划审查过程中标记可能不可接受的射束孔径的有效工具。因此，我们将在临床上部署这种质量验证技术，作为全自动治疗计划工具和自动化计划质量保证计划的一部分。