Patients
This retrospective study was approved by the constituted ethics committee of the Affiliated Hospital of North Sichuan Medical College. And this ethics committee waived the need for informed consent before all patients participated in this study.
From January 2017 to December 2020, we totally collected 489 consecutive patients with biopsy-confirmed thoracic ESCC who underwent MDCT scans. In accordance with the National Comprehensive Cancer Network (NCCN) guidelines [6], the diagnostic criteria for unresectable esophageal cancer on CT were as follows: (a) cT4b tumors with involvement of the trachea, heart, great vessels, or adjacent organs including lung, liver, pancreas and spleen were considered unresectable; (b) ESCC with multi-station, bulky lymphadenopathy was considered unresectable; or (c) ESCC with distant metastases including non-regional lymph nodes (stage IV) was considered unresectable. If ESCC was not considered unresectable based on the criteria, the tumor should be regarded resectable. Among the 489 patients, 342 and 147 had been classified into resectable and unresectable ESCCs, respectively.
Patients with resectable and unresectable ESCC were recruited into our study based on the following inclusion criteria [8]: (a) the patients could be classified as resectable and unresectable ESCCs according to the NCCN guidelines as depicted on MDCT [6]; (b) the patients did not accept any preoperative tumor-related treatments (e.g., radiotherapy or chemotherapy) before receiving CT scans; and (c) the patients did not have distant metastases from ESCC on CT findings. The exclusion criteria in this study were as follows: (a) the quality of CT images was poor (n = 5); (b) the clinicopathological information was incomplete (n = 4); or (c) ESCC was considered resectable on the basis of the NCCN guidelines, but the patients were not able to tolerate general anesthesia and surgery (n = 7). As a result, 16 of 489 patients were excluded, and totally 473 patients were included in our study.
Among the 473 patients, 331 and 142 were classified as resectable and unresectable ESCCs, respectively. In resectable group, the resectability of ESCC was confirmed by histopathological biopsy during operation, and the margin was confirmed to be negative after the operation. Of the 331 patients with resectable ESCC, 20 patients accepted neoadjuvant therapy after CT scans and before surgery, then the tumor size was significantly shrank after therapy, and the cases turned into resectable tumors and underwent surgical treatment, and the resectability was also confirmed by histopathological biopsy during operation. Ultimately, 331 with resectable ESCC and 142 with unresectable cancer were randomly divided into the training cohort (TC, n = 376) and the validation cohort (VC, n = 97). The clinicopathology information in TC and VC are shown in Table 1. This study is schematically presented in Fig. 1.
Table 1
The clinical information of the training and validation cohorts
Variable | The training cohort (n = 376) (Resectable = 263, Unresectable = 113) | The validation cohort (n = 97) (Resectable = 68, Unresectable = 29) |
Median age (years; range) | 66 (42–86) | 65 (49–86) |
Gender (male: female) | 287:89 | 65:32 |
Anatomic distribution (%) | | |
| Upper thoracic portion | 43 (11.5) | 23 (23.7) |
| Middle thoracic portion | 272 (72.3) | 44 (45.4) |
| Lower thoracic portion | 61 (16.2) | 30 (30.9) |
T stage (%) | | |
| cT1 | 56 (14.9) | 16 (16.5) |
| cT2 | 59 (15.7) | 16 (16.5) |
| cT3 | 151 (40.2) | 46 (47.4) |
| cT4a | 25 (6.6) | 4 (4.1) |
| cT4b | 85 (22.6) | 15 (15.5) |
N stage (%) | | |
| N0 | 168 (44.7) | 41 (42.3) |
| N1 | 102 (27.1) | 24 (24.7) |
| N2 | 67 (17.8) | 17 (17.5) |
| N3 | 39 (10.4) | 15 (15.5) |
GTV in cm3 (mean ± SD) | 22.88 ± 20.95 | 22.40 ± 19.26 |
Notes: GTV, gross tumor volume; and SD, standard deviation. |
Contrast-enhanced Computed Tomography
All patients underwent thoracic contrast-enhanced scans with a 64-row MDCT scanner (LightSpeed VCT, GE Medical systems, USA). The interval time between CT and surgery ranged from 2 to 14 days (mean, 8 days). Before CT data acquisition, 100- to 200-mL water was required to drink as esophageal negative contrast material. Following a conventional unenhanced CT scan in the supine position, the contrast-enhanced data acquisition was started 25 to 30 sec after the beginning of contrast material injection (Omnipaque, Iohexol, GE Healthcare, USA) at a rate of 3.0 mL/sec for a total of 70 to 100 mL via a 20-gauge needle inserted into an antecubital vein with an automated injector (Vistron CT Injection System, Medrad, USA). The dosage of injected contrast agent was tailored to body weight at the ratio of 1.5 mL/kg body weight, and then flushed by 20-mL saline. Examinations were executed during one breath hold at full suspended inspiration for 10–15 sec. The parameters of MDCT scanning were as follows: tube voltage of 120 kV, tube current of 200 mA, detector collimation of 64 × 0.6 mm, rotation time of 0.5 sec, pitch of 0.9, slice thickness of 5 mm, and matrix of 512 × 512 mm. The anatomic coverage of CT scan was from the thoracic entrance to the middle level of the left kidney. All the image data were then directly transferred to the General Electric Advantage Workstation 4.4 at the mediastinal window settings (conventional window level, 40 HU; window width, 400 HU).
GTV Measurement
According to the NCCN guidelines [6], the thoracic esophagus was divided into the upper, middle and lower portions through the lower edge of azygos vein and lower edge of lower pulmonary vein. GTV values of resectable and unresectable ESCC based on the involved thoracic portions in TC and VC were measured at the General Electric Advantage Workstation 4.4 at the mediastinal window. In order to obtain GTV, we considered the esophageal wall as abnormal when the thickness exceeded 5 mm on axial image [13]. GTV was calculated by multiplying the sum of all the tumor areas by the layer thickness based on a published method [11, 12]. The circumference of the ESCC was manually depicted along the visible margin of the thickened esophageal wall on each axial contrast-enhanced CT scan to automatically obtain the cross-sectional area of the tumor (Fig. 2a and b). The previous procedure and analysis were repeated on each contiguous axial slice where the tumor was visible. To accurately measure the tumor areas of ESCC, care was taken to avoid the liquid and air in the lumen of the esophagus.
Two radiologists (Reader 1 with 3 years of experience in radiology and Reader 2 with 24 years of experience in radiology) measured the GTV of all non-distant metastatic ESCC patients independently in TC and VC without any knowledge of the histopathological results to test the interobserver reproducibility of the measurement. Before the above CT measurements, Reader 1 was trained in measurement randomly in 20 patients of TC by Reader 2. To verify the intraobserver reproducibility of GTV, measurements of all patients in TC and VC were repeated one month later by Reader 1.
Statistical Analysis
All statistical analyses were carried out by a statistical software (version 26.0 for windows; SPSS, Chicago, IL, USA). The intraclass correlation coefficient (ICC) was used to evaluate the interobserver and intraobserver reliability of repeated measurements of GTV. ICCs less than 0.5, between 0.5 and 0.75, between 0.75 and 0.9, and larger than 0.90 represent poor, moderate, good and excellent reliability, respectively [14].
The univariate and multivariate analyses were performed using the TC dataset. A P-value of less than 0.05 was considered to represent a significant difference. The univariate analysis of possible risk factors of non-distant metastatic ESCC resectability including age, gender, anatomic distribution, cT stage, cN stage and GTV were assessed by using the Chi-square test or Fisher’s exact test in TC. The variables with significant statistical difference in univariate analysis were then enrolled into multivariate analysis, which were carried out by the binary logistic regression analysis to determine the independent risk factors. Subsequently, the Mann-Whitney U test was applied to compare GTV between patients of resectable and unresectable ESCC corresponding to different anatomic distributions. If a significant difference was demonstrated, the receiver operating characteristic (ROC) analysis was performed to ascertain whether the cutoff values of GTV based on anatomic distributions could help determinate the resectability.
To evaluate the agreements between VC and TC, unweighted Cohen’s Kappa tests were used to validate the performance of the previous ROC models for the resectability of non-distant metastatic ESCC in the VC dataset independently [15]. We applied unweighted Cohen’s Kappa test based on the following rating scale: less than 0.20, poor agreement; 0.21 to 0.40, fair agreement; 0.41 to 0.60, moderate agreement; 0.61 to 0.80, good agreement; and more than 0.81, excellent agreement [16].