Costea Madalina, Zlate Alexandra, Serre Anne-Agathe, Racadot Séverine, Baudier Thomas, Chabaud Sylvie, Grégoire Vincent, Sarrut David, Biston Marie-Claude
Centre Léon Bérard, 28 rue Laennec, LYON 69373 Cedex 08, France; CREATIS, CNRS UMR5220, Inserm U1044, INSA-Lyon, Université Lyon 1, Villeurbanne, France.
MedEuropa, Strada Turnului 8, Brașov 500152, Romania.
Radiother Oncol. 2023 Nov;188:109870. doi: 10.1016/j.radonc.2023.109870. Epub 2023 Aug 25.
To investigate the performance of 4 atlas-based (multi-ABAS) and 2 deep learning (DL) solutions for head-and-neck (HN) elective nodes (CTVn) automatic segmentation (AS) on CT images.
Bilateral CTVn levels of 69 HN cancer patients were delineated on contrast-enhanced planning CT. Ten and 49 patients were used for atlas library and for training a mono-centric DL model, respectively. The remaining 20 patients were used for testing. Additionally, three commercial multi-ABAS methods and one commercial multi-centric DL solution were investigated. Quantitative evaluation was assessed using volumetric Dice Similarity Coefficient (DSC) and 95-percentile Hausdorff distance (HD). Blind evaluation was performed for 3 solutions by 4 physicians. One recorded the time needed for manual corrections. A dosimetric study was finally conducted using automated planning.
Overall DL solutions had better DSC and HD results than multi-ABAS methods. No statistically significant difference was found between the 2 DL solutions. However, the contours provided by multi-centric DL solution were preferred by all physicians and were also faster to correct (1.1 min vs 4.17 min, on average). Manual corrections for multi-ABAS contours took on average 6.52 min Overall, decreased contour accuracy was observed from CTVn2 to CTVn3 and to CTVn4. Using the AS contours in treatment planning resulted in underdosage of the elective target volume.
Among all methods, the multi-centric DL method showed the highest delineation accuracy and was better rated by experts. Manual corrections remain necessary to avoid elective target underdosage. Finally, AS contours help reducing the workload of manual delineation task.
研究4种基于图谱（多图谱自动分割系统，multi-ABAS）和2种深度学习（DL）方法对头颈部（HN）选择性淋巴结（CTVn）在CT图像上进行自动分割（AS）的性能。
在增强扫描的计划CT上勾画出69例HN癌症患者双侧CTVn的范围。分别将10例和49例患者用于图谱库构建和训练单中心DL模型。其余20例患者用于测试。此外，还研究了三种商用多图谱自动分割方法和一种商用多中心DL解决方案。使用体积骰子相似系数（DSC）和95%百分位数豪斯多夫距离（HD）进行定量评估。由4名医生对3种解决方案进行盲法评估。其中一人记录手动校正所需时间。最后使用自动计划进行剂量学研究。
总体而言，DL解决方案的DSC和HD结果优于多图谱自动分割方法。两种DL解决方案之间未发现统计学显著差异。然而，所有医生都更喜欢多中心DL解决方案提供的轮廓，并且校正速度也更快（平均1.1分钟对4.17分钟）。多图谱自动分割轮廓的手动校正平均耗时6.52分钟。总体而言，从CTVn2到CTVn3再到CTVn4，轮廓准确性呈下降趋势。在治疗计划中使用自动分割轮廓会导致选择性靶区剂量不足。
在所有方法中，多中心DL方法显示出最高的勾画准确性，并且得到专家的更高评价。仍需要进行手动校正以避免选择性靶区剂量不足。最后，自动分割轮廓有助于减少手动勾画任务的工作量。
