CT-RTSTRUCT-RTDOSE-RTPLAN Sets of Head and Neck Cancers Treated with Identical Prescriptions using IMRT: An Open Dataset for Deep Learning in Treatment Planning

<p>This collection includes data from 211 patients who presented with head and neck cancer and were treated using radiation therapy at a single institution using 6 MV Intensity Modulated Radiation Therapy (IMRT) from a linear accelerator. All patients were prescribed identical prescriptions of 70 Gray (Gy) in 33 fractions for the primary planning target volume (PTV) and one, none, or a combination of integrated boost sub targets (PTV 54 Gy, PTV 56 Gy, PTV 57 Gy, PTV 60 Gy, PTV 63 Gy, PTV 66 Gy).  The data for each patient contains the minimum required information for radiation treatment planning. Each patient set contains the planning computed tomography (CT) image set from treatment simulation, the expert-defined radiotherapy structure set (RTSTRUCT), the delivered radiotherapy plan file (RTPLAN), and the calculated treatment dose (RTDOSE). All files are provided in DICOM format. </p><p><u>CT Images:</u> Planning CT images from treatment simulation and radiation dose calculation are included for each patient. Thermoplastic head and neck masks were used during image acquisition to achieve immobilization and ensure accurate and reproducible positioning during treatment. All images satisfied the resolution requirements for dose calculations and have an identical slice thicknesses of 2.5 mm. The provided CT files are those that match with the RTDOSE and RTSTRUCT files of the same patient. </p><p><u>Target and Organ Contours:</u> Each patient’s data contains an RTSTRUCT DICOM with manually defined target volume contours and organ at risk (OAR) contours. For efficient OAR extraction from the RTSTRUCT files, we renamed all OAR contours that could be identified to a uniform structure name. If a contour was not identified, it was not defined for treatment, or the nomenclature was ambiguous. All planning target volumes (PTV) are included in the structure set. Due to inconsistent documentation and naming conventions, the primary PTV is difficult to distinguish from other PTV structures. For example, some patients include multiple iterations of the same contour as a result of modifications during treatment planning our assisted planning methods from the TPS. In addition to organ and target structures, we chose to leave in planning structures created and used by the dosimetrists for generating the plan. For example, PTV expansions, ring structures, avoidance structures, and normal tissue volumes were left within the RTSTRUCT file. OAR contours for 26 unique normal tissue structures were renamed to the <a href="/wp-content/uploads/Standardized-OAR-Names.csv">‘Standardized Name</a>’ in the RTSTRUCT DICOM file. If a structure was not identified, the structure was not created in the original structure set. </p><p><u>Radiotherapy Dose and Treatment Plan:</u> Each dose was planned by a dosimetrist and approved for treatment by the physician. Treatment plans were created using Pinnacle (Phillips Medical Systems, Fitchburg, WI). The dose was calculated on a 3 mm3 dose grid and plans were optimized for treatment on Elekta or Varian linear accelerators.</p><p>Research for challenging tumor sites like head and neck cancers can benefit from publicly accessible radiotherapy treatment datasets such as the one offered here. In particular, this dataset provides consistent treatment plans with limited variance due to the uniformity in tumor site, consistent institutional standards, and identical prescriptions. In artificial intelligence research, the need for large and uniform datasets is crucial for testing and evaluating the performance of novel architectures. Treatments of a single site from a single institution and with identical prescriptions will have reduced variance of tumor coverage and normal tissue-sparing objectives in comparison to multi-site and multi-institutional studies. However, it can also be combined with other radiotherapy treatment datasets for use in more generalized studies.</p>

本数据集包含来自211名患者的数据，这些患者因头颈部癌症在接受单一机构采用6 MV调强放射治疗（IMRT）治疗。所有患者均被开具了相同的处方，即对主要计划靶体积（PTV）进行33次分剂量的70 Gray（Gy）照射，并针对一个、零个或多个综合增敏亚靶区（PTV 54 Gy、PTV 56 Gy、PTV 57 Gy、PTV 60 Gy、PTV 63 Gy、PTV 66 Gy）进行照射。每位患者的数据包含了放射治疗计划所需的最基本信息。每个患者集包含治疗模拟的规划计算机断层扫描（CT）图像集、专家定义的放射治疗结构集（RTSTRUCT）、已实施的放射治疗计划文件（RTPLAN）以及计算出的治疗剂量（RTDOSE）。所有文件均以DICOM格式提供。

underline{CT图像：}为每位患者提供了治疗模拟和辐射剂量计算的规划CT图像。在图像采集过程中使用了热塑性头颈部面具以实现固定，并确保治疗过程中的准确和可重复定位。所有图像均满足剂量计算的分辨率要求，并具有相同的2.5毫米切片厚度。提供的CT文件与相同患者的RTDOSE和RTSTRUCT文件匹配。

underline{靶区和器官轮廓：}每位患者的数据包含一个手动定义的靶体积轮廓和风险器官（OAR）轮廓的RTSTRUCT DICOM文件。为了从RTSTRUCT文件中高效提取OAR，我们将所有可识别的OAR轮廓重命名为统一的结构名称。如果轮廓未被识别，则未对其进行治疗定义，或命名学不明确。所有计划靶体积（PTV）均包含在结构集中。由于文档记录和命名规范的不一致，主PTV难以与其他PTV结构区分。例如，一些患者因治疗计划过程中的修改或来自TPS辅助计划方法的多重迭代而包含多个相同的轮廓。此外，我们还选择保留由剂量学家创建并用于生成计划的规划结构。例如，PTV扩展、环形结构、避障结构和正常组织体积等均保留在RTSTRUCT文件中。26个独特的正常组织结构的OAR轮廓在RTSTRUCT DICOM文件中重命名为‘标准化名称’。如果结构未被识别，则原始结构集中未创建该结构。

underline{放射治疗剂量和治疗计划：}每位剂量均由剂量学家计划并由医师批准进行治疗。治疗计划使用Pinnacle（菲利普斯医疗系统，威斯康星州菲奇堡）创建。剂量在3 mm3剂量网格上计算，计划针对Elekta或Varian直线加速器进行治疗优化。

研究具有挑战性的肿瘤部位，如头颈部癌症，可以从公开可访问的放射治疗治疗数据集中受益，例如此处提供的此类数据集。特别是，该数据集提供了由于肿瘤部位、机构标准一致性和处方相同而导致的有限变异的治疗计划。与多地点和多机构研究相比，来自单一机构单一地点且处方相同的治疗将具有降低肿瘤覆盖率和正常组织保护目标的变异。然而，它也可以与其他放射治疗治疗数据集结合，用于更广泛的研究。

搜集汇总

背景与挑战

背景概述

该数据集是一个开放数据集，包含211名头颈癌患者的放疗数据，所有患者均采用相同的IMRT处方剂量（70 Gy分33次），数据包括计划CT、RTSTRUCT、RTPLAN和RTDOSE文件，以DICOM格式提供。其特点在于数据高度统一，减少了肿瘤覆盖和正常组织保护的方差，适用于深度学习在放疗计划中的应用，尤其适合测试和评估新算法模型。