IEEE J Biomed Health Inform. 2021 Mar;25(3):755-763. doi: 10.1109/JBHI.2020.3002805. Epub 2021 Mar 5.
Radiomics, an emerging tool for medical image analysis, is potential towards precisely characterizing gastric cancer (GC). Whether using one-slice 2D annotation or whole-volume 3D annotation remains a long-time debate, especially for heterogeneous GC. We comprehensively compared 2D and 3D radiomic features' representation and discrimination capacity regarding GC, via three tasks ( T, lymph node metastasis' prediction; T, lymphovascular invasion's prediction; T, pT4 or other pT stages' classification).
Four-center 539 GC patients were retrospectively enrolled and divided into the training and validation cohorts. From 2D or 3D regions of interest (ROIs) annotated by radiologists, radiomic features were extracted respectively. Feature selection and model construction procedures were customed for each combination of two modalities (2D or 3D) and three tasks. Subsequently, six machine learning models ( Model, Model; Model, Model; Model, Model) were derived and evaluated to reflect modalities' performances in characterizing GC. Furthermore, we performed an auxiliary experiment to assess modalities' performances when resampling spacing different.
Regarding three tasks, the yielded areas under the curve (AUCs) were: Model's 0.712 (95% confidence interval, 0.613-0.811), Model's 0.680 (0.584-0.775); Model's 0.677 (0.595-0.761), Model's 0.615 (0.528-0.703); Model's 0.840 (0.779-0.901), Model's 0.813 (0.747-0.879). Moreover, the auxiliary experiment indicated that Models are statistically advantageous than Models with different resampling spacings.
Models constructed with 2D radiomic features revealed comparable performances with those constructed with 3D features in characterizing GC.
Our work indicated that time-saving 2D annotation would be the better choice in GC, and provided a related reference to further radiomics-based researches.
放射组学是一种新兴的医学图像分析工具，有望精确地对胃癌（GC）进行特征描述。在对 GC 进行分析时，使用单层面 2D 注释还是全容积 3D 注释一直存在争议，尤其是对于异质性 GC。我们通过三个任务（T1，淋巴结转移预测；T1，血管淋巴管侵犯预测；T4 期或其他 T 分期分类），全面比较了 2D 和 3D 放射组学特征对 GC 的表示和区分能力。
回顾性纳入了来自四个中心的 539 名 GC 患者，并将其分为训练集和验证集。由放射科医生分别对 2D 或 3D 感兴趣区（ROI）进行注释，提取放射组学特征。为每种两种模式（2D 或 3D）和三个任务的组合定制了特征选择和模型构建程序。随后，衍生并评估了六个机器学习模型（Model1，Model2；Model3，Model4；Model5，Model6），以反映两种模式在描述 GC 方面的性能。此外，我们进行了一个辅助实验，以评估在不同采样间隔下两种模式的性能。
在三个任务中，曲线下面积（AUC）的结果为：Model1 的 0.712（95%置信区间，0.613-0.811），Model2 的 0.680（0.584-0.775）；Model3 的 0.677（0.595-0.761），Model4 的 0.615（0.528-0.703）；Model5 的 0.840（0.779-0.901），Model6 的 0.813（0.747-0.879）。此外，辅助实验表明，在不同的采样间隔下，Model1 比 Model2 具有统计学优势。
用 2D 放射组学特征构建的模型在描述 GC 方面表现出与用 3D 特征构建的模型相当的性能。
我们的工作表明，在 GC 中，节省时间的 2D 注释将是更好的选择，并为进一步的基于放射组学的研究提供了相关参考。
