Gross Tumor Volume Definition and Comparative Assessment for Esophageal Squamous Cell Carcinoma From 3D 18F-FDG PET/CT by Deep Learning-Based Method

Abstract

Background: The accurate definition of gross tumor volume (GTV) of esophageal squamous cell carcinoma (ESCC) can promote precise irradiation field determination, and further achieve the radiotherapy curative effect. This retrospective study is intended to assess the applicability of leveraging deep learning-based method to automatically define the GTV from 3D ¹⁸F-FDG PET/CT images of patients diagnosed with ESCC.

Methods: We perform experiments on a clinical cohort with 164 ¹⁸F-FDG PET/CT scans. The state-of-the-art esophageal GTV segmentation deep neural net is first employed to delineate the lesion area on PET/CT images. Afterwards, we propose a novel equivalent truncated elliptical cone integral method (ETECIM) to estimate the GTV value. Indexes of Dice similarity coefficient (DSC), Hausdorff distance (HD), and mean surface distance (MSD) are used to evaluate the segmentation performance. Conformity index (CI), degree of inclusion (DI), and motion vector (MV) are used to assess the differences between predicted and ground truth tumors. Statistical differences in the GTV, DI, and position are also determined.

Results: We perform 4-fold cross-validation for evaluation, reporting the values of DSC, HD, and MSD as 0.72 ± 0.02, 11.87 ± 4.20 mm, and 2.43 ± 0.60 mm (mean ± standard deviation), respectively. Pearson correlations (R²) achieve 0.8434, 0.8004, 0.9239, and 0.7119 for each fold cross-validation, and there is no significant difference (t = 1.193, p = 0.235) between the predicted and ground truth GTVs. For DI, a significant difference is found (t = -2.263, p = 0.009). For position assessment, there is no significant difference (left-right in x direction: t = 0.102, p = 0.919, anterior-posterior in y direction: t = 0.221, p = 0.826, and cranial-caudal in z direction: t = 0.569, p = 0.570) between the predicted and ground truth GTVs. The median of CI is 0.63, and the gotten MV is small.

Conclusions: The predicted tumors correspond well with the manual ground truth. The proposed GTV estimation approach ETECIM is more precise than the most commonly used voxel volume summation method. The ground truth GTVs can be solved out due to the good linear correlation with the predicted results. Deep learning-based method shows its promising in GTV definition and clinical radiotherapy application.

Keywords: comparative assessment; deep learning; definition; equivalent truncated elliptical cone; esophageal squamous cell carcinoma; gross tumor volume.

Figure 1

Two existing methods for GTV estimation. (A) A cross-section view of the voxel volume summation method. The middle white part denotes the lesion mask, and the surrounding dash area denotes the extra computation. (B) Two opposed truncated cones summation method, d ₁ and d ₃ respectively represent the cranial and caudal transverse diameters of the tumor. d ₂ is the maximal transverse diameter, and h is the tumor height.

Figure 2

Esophageal carcinoma approximately assessed by its corresponding equivalent ellipses. CT_i and GroTr_i (i = 1,2…, 10) denote the i th CT slice and its corresponding ground truth mask of tumor. The red ellipse is the equivalent ellipse of lesion. The white mask is the lesion mask. The intersection angles between the green line segment and the blue horizontal straightness represent the sharp angles between the long principal axes and the positive X-axis.

Figure 3

Overview. The whole GTV definition process for ESCC patient includes four stages: data acquisition, data preprocessing, segmentation, and GTV estimation.

Figure 4

Segmentation visual results. The more slices in patients (A) denote larger tumor than (B). The red contours are the predicted results by PSNN, and the blue contours represent the ground truth.

Figure 5

Results of GTV assessment. (A) Scatter plot and correlation between the predicted GTVs by ETECIM and manual ground truth GTVs, for the first fold cross-validation. (B) Scatter plot and correlation between the predicted GTVs by voxel summation method and manual ground truth GTVs, for the first fold cross-validation. For the same reason, (C, D) are results for the second fold cross-validation. (E, F) are results for the third fold cross-validation. (G, H) are results for the fourth fold cross-validation.