Selection of Patients and Their Characteristics
This retrospective study was approved by the Institutional Review Boards of the Affiliated Hospital of Qingdao University (institution 1) and the Shandong Provincial Hospital affiliated to Shandong First Medical University (institution 2), which waived patient informed consent. We screened the patients with pathologically confirmed GISTs in the databases of these institutions from July 2008 to August 2019. Patients who fulfilled the following inclusion criteria were included in the study: (a) definitive genetic analysis results could be obtained; (b) complete clinical characteristics could be obtained; (c) tri-phase CE-CT images before treatment (neoadjuvant therapy or surgery) could be obtained; and (d) the GIST size was ≥ 2 cm and could be outlined in the CE-CT images for volume of interest segmentation.
According to these criteria, 106 patients who also underwent genetic analysis between August 2010 and August 2019 were included: 93 were from institution 1 and 13 were from institution 2. Because the number of patients from institution 2 was small, their data could not be used as an external validation set alone. Therefore, the data of patients from both institutions were merged, then randomly allocated to training and validation sets in a 3:1 ratio. The number of patients with the KIT exon 11 mutation was 45, the number of patients without the KIT exon 11 mutation was 61. Among the 106 patients, 57 were male and 49 were female (mean age 59.5 ± 9.9 years). Detailed information on the clinical characteristics of the selected patients is given in Table 1. A detailed description of the workflow process is shown in Fig. 1.
Table 1
Clinicopathologic and Demographic Data
Clinical factors
|
KIT exon 11 mutation (-), n = 45
|
KIT exon 11 mutation (+), n = 61
|
P value
|
Gender
|
Male
|
22
|
35
|
0.386
|
Female
|
23
|
26
|
Age (Median[range]), year
|
|
57 (47.50–66.50)
|
62 (55.00–66.00)
|
0.051
|
Growth pattern
|
Endoluminal
|
12
|
20
|
0.238
|
Exophytic
|
15
|
26
|
Mixed
|
18
|
15
|
CEA(Median[range]), ng/ml
|
|
1.76 (1.14–2.91)
|
2.00 (1.18–3.16)
|
0.623
|
CA199(Median[range]), ng/ml
|
|
10.61 (6.22–17.07)
|
8.64 (5.66–13.82)
|
0.215
|
CA125(Median[range]), ng/ml
|
|
10.06 (7.56–27.23)
|
9.06 (7.55–11.52)
|
0.144
|
CA724(Median[range]), ng/ml
|
|
4.40 (1.24–17.59)
|
1.49 (1.07–3.91)
|
0.006
|
Maximum diameter of tumor
|
|
5.00 (3.50-8.00)
|
5.00 (3.85–8.85)
|
0.468
|
Mitotic count (mitoses/50 hpf)
|
< 5
|
31
|
32
|
0.089
|
≥ 5
|
14
|
29
|
Anatomic location
|
Stomach
|
21
|
38
|
0.109
|
Small bowel
|
18
|
20
|
Colorectum
|
2
|
2
|
Mesentery
|
4
|
1
|
Ki-67(Median[range]), %
|
|
5 (3–11)
|
10 (5–20)
|
0.070
|
CD-117
|
(-)
|
5
|
0
|
0.012
|
(+)
|
40
|
61
|
CD-34
|
(-)
|
9
|
7
|
0.226
|
(+)
|
36
|
54
|
DOG-1
|
(-)
|
4
|
1
|
0.160
|
(+)
|
41
|
60
|
NIH risk stratification (2008)
|
Very low
|
2
|
2
|
0.444
|
Low
|
14
|
16
|
Intermediate
|
10
|
13
|
High
|
19
|
30
|
Note: NIH, National Institutes of Health; CEA, carcinoembryonic antigen; CA199, carbohydrate antigen 199; CA125, carbohydrate antigen 125; CA724, carbohydrate antigen 724; (+), positive ;(-), negative. |
Tri-phase CE-CT Image Screening and Volume of Interest Segmentation
All 106 patient had undergone CE-CT scanning with 16 or 64-detector spiral CT scanners (Siemens Somatom Definition, GE Hispeed, GE Bright, or GE Optima 670). For all these measurements, the tube current was 200–240 mAs, the slice thickness was 5.0 mm, the tube voltage was 120–160 kV, and the injection rate was 3.5 ml/s, with intravenous injection of contrast (iopromide, 80 ml) and arterial, venous, and delayed phase with 30, 70, and 300 s delay, respectively.
ITK-SNAP (v3.6.0; www.itksnap.org) was used for three-dimensional volume of interest (VOI) segmentation of the CE-CT images. Radiologist 1, who had nine years’ clinical experience of using ITK-SNAP, outlined the edge of the tumor on the tri-phase CE-CT images layer by layer and fused them into the VOI. The same radiologist repeated the VOI segmentation and feature extraction two weeks later. In addition, radiologist 2, who had 14 years’ clinical experience of using ITK-SNAP, performed the VOI segmentation and feature extraction once. To evaluate inter- and intra-observer reproducibility, the two radiologists randomly selected 30 cases for VOI segmentation and feature extraction. Inter-/ intra-class correlation coefficients (ICCs) were used to evaluate the reproducibility and stability of radiomics feature extraction to identify robust radiomics features. The inter-observer ICC was derived from the first feature extraction by radiologist 1 and the feature extraction by radiologist 2. The intra-observer ICC was derived from the feature extraction performed twice by radiologist 1. Generally, ICCs > 0.75 are considered to indicate good reproducibility or reliability. In this study, to ensure highly accurate features were selected, an ICC > 0.9 was used as a criterion for good robustness.
Image Normalization and Feature Extraction
For the CE-CT regions, the µ ± 3 σ method (18), gray-level quantization (19), and cubic interpolation (voxel size = 1 × 1 × 1 mm3) (20) were used to redress the influences of distinct acquisition protocols, scales, and directions.
All the CE-CT images were processed using AnalysisKit software (v1.0.3; GE Healthcare, China) for texture feature extraction and quantification analysis of tissue heterogeneity. A total of 396 textural parameters were extracted, including the characteristics of 42 histograms, 10 Haralick features, 9 form factors, 48 grey-level co-occurrence matrix (GLCM) features with an offset of 1/4/7, and 60 gray-level run length matrix (GLRLM) features with an offset of 1/4/7.
Combat Compensation Method
The distinguishing features of the texture patterns were retained using the combat compensation method (21), while eliminating the influence of the scanner and the protocol. This method was helpful for multi-center radiomics analysis (22) and can be used for CE-CT images.
Clinical Characteristics
The preoperative demographic and clinicopathologic data, including age, sex, and blood test results, were collected for the 106 patients. Postoperative data, including the location and size (maximum diameter) of the tumor, proteomics (e.g., expression statuses of tumor related proteins Ki-67, CD117, CD34, and DOG1), mitotic count, National Institutes of Health (NIH) risk stratification, growth pattern, and genetic phenotype, also were collected.
Clinical Model, Radiomics Algorithm, and Radiomics Nomogram Building
The relationship between clinical factors and the KIT exon 11 mutation was assessed by univariate logistic regression. Then, a multivariate logistic model was used with the significant clinical features (P < 0.05) to develop the clinical model.
The feature extraction algorithm implemented in the R statistical software (v3.6.3; www.Rproject.org) was used to select features and generate the models. To select the most valuable predictive factors in the training set, we incorporated the top 20 features of minimum redundancy maximum relevance (mRMR) into the least absolute shrinkage and selection operator method (LASSO) and fitted for regression of data. Generalized linear model (GLM), which is a machine-learning classifier, was trained with the training set using the 10-fold cross-validation approach. A radiomics algorithm was added with preoperative and postoperative clinical factors to build predictive nomograms for the training set with multivariable logistic regression analysis.
Assessment and Validation of the Clinical Model, Radiomics Algorithm, and Radiomics Nomogram
The effectiveness of the clinical model, radiomics algorithm, and radiomics nomogram in differentiating, calibration, and clinical value was verified with the validation set. The predictive capability was evaluated using the validation set by the receiver operating characteristic (ROC) curve, the area under the ROC curve (AUC), specificity, sensitivity, accuracy, and positive and negative predictive values. The calibration curves were used to assess the agreement between the predicted and actual results with the Hosmer–Lemeshow test (23). The clinical applicability of the radiomics algorithm and preoperative and postoperative radiomics nomograms also was assessed with decision curve analysis by calculating the net benefit at different threshold probabilities (24).
Statistical Analysis
R statistical software (v3.6.3; www.Rproject.org) was used for all the statistical analyses.
To select the clinical factors, Fisher’s exact test or the chi-square test was used to compare the categorical variables between the training and validation sets. The Mann-Whitney U test or independent t-test was used to evaluate the discrimination in clinical characteristics among different groups of continuous variables. The relationship between the KIT exon 11 mutation and clinical factors was assessed with univariate analysis.