Prediction of Ki-67 expression and malignant potential in gastrointestinal stromal tumors: novel models based on CE-CT and serological indicators

doi:10.21203/rs.3.rs-5094652/v1

Background and study aims To identify more reliable imaging and serological indicators for predicting Ki-67 expression and malignant potential in gastrointestinal stromal tumors, as well as to develop a preoperative prediction model with clinical utility.

Patients and methods Patients with gastrointestinal stromal tumors diagnosed at the First Affiliated Hospital of Jinzhou Medical University between May 2018 and May 2024 were retrospectively analyzed. Univariate logistic analyses, multivariate logistic analyses, stepwise regression analyses, and LASSO regression analyses were utilized to identify Ki-67 high expression and high malignant potential risk factors for GIST. Prediction models were then constructed and nomograms developed.

Results Tumor diameter and EVFDM were found to be closely associated with Ki-67 expression, while tumor volume and IBSC were closely related to malignant potential. The two-way stepwise regression model demonstrated high accuracy and good fit. The AUCs for the Ki-67 expression model in the training and validation groups were 0.865 (95%CI 0.807-0.922) and 0.784 (95%CI 0.631-0.937) respectively. The AUCs for the malignant potential model in the training and validation groups were 0.950 (95%CI 0.920-0.980) and 0.936 (95%CI 0.867-1.000) respectively.

Conclusion The maximum diameter, growth pattern, EVFDM, peritumoral vessels, PLR, OPNI are correlated with Ki-67 high expression. Volume, contour, ulcer, IBSC and OPNI are correlated with malignant potential of gastrointestinal stromal tumors; Preoperative models developed using enhanced CT imaging can respectively predict the malignant potential and Ki-67 expression of GIST. Furthermore, when combined with serological indicators, the predictive accuracy of this model can be further enhanced.

Gastrointestinal stromal tumor

Contrast-enhanced CT

Serological indicators

Ki-67

malignant potential

Model

Gastrointestinal stromal tumor (GIST) is the most common stromal tumor of the digestive tract, typically originating from the muscularis propria and stemming from the interstitial cells of Cajal (ICC), which are pacemaker cells of the gastrointestinal tract^[1]. Considered a potentially malignant sarcoma, GIST is often evaluated using the modified NIH risk classification criteria, which take into account factors such as tumor size, mitotic rate, location, and rupture. This classification categorizes GIST recurrence risk into very low, low, intermediate, and high risk groups^[2].The current classification standards for GIST are primarily based on postoperative pathology, but there is a need for clearer preoperative evaluation criteria. Preoperative risk assessment plays a crucial role in determining treatment options and predicting prognosis, surgery is an effective method for treating localized Gastrointestinal Stromal Tumors (GIST), while imatinib is a crucial treatment option for metastatic GIST and high-risk GIST cases ^[3] .In order to select a more appropriate treatment plan, some scholars have developed prediction models for GIST Ki-67 expression and malignant potential using imaging techniques^༻4–7༽, few studies have utilized serological indicators in conjunction with imaging features to predict the malignant potential of GISTs. This study aimed to develop and validate a comprehensive model for predicting the malignant potential of GIST using imaging characteristics and serological indicators.

Ki-67 can be used as an indicator of tumor proliferation. Tumors with stronger proliferation abilities tend to have higher malignant potential^[8]. Some smaller tumors are incidentally discovered during examinations, and relying solely on their size to evaluate their malignant potential and proliferation ability may be unreliable. For instance, even small GISTs measuring less than 2cm may have malignant potential and the ability to metastasize^[9]. The malignant potential of GIST may not align entirely with Ki-67 expression due to various factors. Firstly, the puncture site can impact results, as the chosen site may not accurately reflect the overall tumor proliferation^[8]. Additionally, when evaluating GIST malignancy, tumor size and mitotic figures are crucial indicators, excluding Ki-67 expression status. This exclusion may underestimate the malignant potential of certain early-stage tumors. Mitotic count only captures cell proliferation in the M phase and does not provide a comprehensive assessment of tumor cells. Conversely, Ki-67 expression increases from the G1 phase to mitosis, with a sudden decrease in the G0 phase^[3].Therefore, some scholars have suggested that Ki-67 has a more comprehensive impact compared to mitotic figures, high expression of Ki-67 is considered an independent risk factor for high malignant potential^[10,11].It is also associated with the selection of treatment options for patients, postoperative survival, and recurrence rates^[12–15]. In order to enhance the comprehensiveness of the preoperative assessment, this study included the prediction of Ki-67.

Patients

A total of 193 patients diagnosed with GIST through postoperative pathology at the First Affiliated Hospital of Jinzhou Medical University between May 2018 and May 2024, and with complete preoperative examination data, were included in this study. The inclusion criteria were: ① age ≥ 18 years; ② Patients with pathologically confirmed GIST after surgical or endoscopic resection; ③ Those with complete preoperative enhanced CT, serological data, postoperative pathological diagnosis, and immunohistochemistry data; ④ Patients who underwent CT examination without tumor rupture or metastasis before surgery and did not receive drug treatment prior to CT examination and surgery. The exclusion criteria were: ① Patients with other tumors, such as gastric cancer, colon cancer, etc.; ② Patients with hematological diseases or immune system diseases, such as leukemia, psoriasis, ankylosing spondylitis, etc.; ③ Patients with nutritional disorders, such as hypoalbuminemia, requiring albumin supplementation before examination; ④ Those with acute or severe infection within one week before surgery; ⑤ Cases where imaging was unclear and patient outlines were difficult to discern. The study included 193 patients, with an average age of 62.38 ± 10.60 years, comprising 85 males and 108 females. This study was approved by the Ethics Review Committee of the First Affiliated Hospital of Jinzhou Medical University, with ethics number KYLL202445.A flowchart of the study population is shown in Fig. 1.

Histopathologic examination

Ki-67 detection and pathological diagnosis were conducted by two pathologists using a 50x microscope. A total of 1,000 cells were randomly selected from each slice, and the Ki-67 positive cells were counted to determine the expression level, There remains ongoing debate regarding the cutoff point for Ki-67 expression levels. While some studies utilize thresholds of 5%, 8%, and 10% ^[5,11], the 10% threshold is generally more accepted. Consequently, this study adopts the 10% cutoff to categorize Ki-67 expression levels as high, dividing them into two groups: high expression (Ki-67 > 10%) and low expression (Ki-67 ≤ 10%). Specimens were histologically examined by pathologists with over 5 years of experience. Tumor specimens with a diameter of ≥ 2cm were cut every 1cm and fully fixed to ensure accuracy for subsequent immunohistochemistry. Postoperative pathological specimens were stained with HE to document tumor size and mitotic figures (recorded in units of 50 high-power fields), and the GIST risk classification was evaluated based on NIH standards.

Serological indicators

The preoperative neutrophil count, monocyte count, platelet count, lymphocyte count, and albumin levels of the patient were retrieved from the hospital's clinical record system. Subsequently, the patient's Neutrophil-to-Lymphocyte Ratio (NLR), Monocyte-to-Lymphocyte Ratio (MLR), Platelet-to-Lymphocyte Ratio (PLR), and Overall Prognostic Nutritional Index (OPNI) were calculated. NLR represents the ratio of neutrophil count to lymphocyte count, PLR represents the ratio of platelet count to lymphocyte count, MLR represents the ratio of monocyte count to lymphocyte count, and OPNI is calculated as the total number of lymphocytes plus 5 times albumin. Cutoff values were utilized to categorize these serological indicators.

CT imaging acquisition

All patients underwent abdominal plain scan and enhancement before surgery. The patients were scanned with a Philips CT scanner from the top of the diaphragm to the level of the lower edge of the pubic symphysis in a single breath-hold scan. The scanning conditions included 120kV and 300mA, with a matrix size of 512x512. Following the non-contrast scan (0s), an automatic injector was used by the examiner to inject 350 mg/ml ioversol into the patient's cubital vein at a speed of 3.0 ~ 5.0ml/s. The dosage was calculated by multiplying the patient's weight by 1.5ml/kg. Subsequently, arterial, venous, and delayed phase images were captured at 25–30 seconds, 65–70 seconds, and 110–120 seconds after contrast injection.

Image analysis

Consistent manual segmentation of tumor areas was conducted on venous phase CT images with a slice thickness of 0.9 mm by two experienced radiologists. A region of interest (ROI) was delineated layer by layer along the tumor contour boundary, covering the entire tumor volume. Two radiologists, one with 20 years and the other with 5 years of imaging experience, independently reviewed the image archives and documented CT image features, such as location (gastric, non-gastric), tumor diameter (cm), volume (cm³), outline (regular, irregular), growth pattern (intraluminal, extraluminal, mixed type), enhancement mode (uniform, uneven), internal components (ulcers, calcifications, necrosis), and tumor blood vessels ( enlarged vessels feeding or draining the mass (EVFDM), intratumoral blood supply conditions(IBSC)). Presence of ulcerations was determined by the observation of a focal tissue defect on the endoluminal surface of the lesion. Lesions classified as target showed hypo-attenuating foci with CT attenuation values ranging from 0 to 20 HU, lacked enhancement post-re-injection, had uneven density, and displayed an irregular inner edge. These characteristics were indicative of intratumoral necrosis. Additionally, target lesions with high-attenuating foci on unenhanced CT scans, which had a CT attenuation value exceeding 90 HU, were categorized as tumor calcification. EVFDM included peritumoral enlarged vessels feeding the mass, which was determined at portal venous phase.Rich blood supply was defined as irregular blood vessels flowing from the edge to the center, while poor blood supply was defined as blood vessels running only at the edge of the tumor. The 3D-slicer software was utilized for outlining the tumor volume and measuring the maximum diameter, and the U-viewer software UCLA plug-in was employed to assess the blood supply within the tumor on arterial phase images. In case of disagreement, it is up to the two to negotiate and decide. The delineation and pathological results of the tumor are shown in Fig. 8.

Statistical analysis

A total of 193 patients were included in the study, and they were randomly divided into a training group and a validation group at a ratio of 8:2. Statistical analysis was conducted using SPSS version 26.0. Continuous variable analysis utilized the independent t-test and Mann-Whitney U test. Normally distributed continuous variables were presented as mean ± standard deviation (X ± S), while skewed distributed variables were expressed as median and interquartile range (IQR). Categorical data were analyzed using chi-square test or Fisher's exact test, with results presented as frequency and percentage (n, %). The optimal cutoff values for continuous variables were determined using receiver operating characteristic (ROC) curve analysis and the Youden index. Interobserver agreement was evaluated using the intraclass correlation coefficient (ICC). Binary logistic regression and stepwise regression were performed using Zstats v1.0, while Lasso regression was conducted using the Hiplot website^[16]. Results were reported as odds ratio (OR) and 95% confidence intervals (CI). The influencing factors were identified and a prediction model was developed through the three regression methods, stepwise regression and Lasso regression utilize the AIC criterion to select independent variables, while binary logistic regression employs a screening method based on a significance level of P value < 0.05 for independent variables. Model comparison was done using the Hosmer-Lemeshow test, Akaike information criterion (AIC), and Bayesian information criterion (BIC) to assess model fit.The area under the curve (AUC) was utilized to assess the accuracy of the prediction model. The optimal model was chosen, and visual representations such as the nomogram, calibration curve, and clinical decision curve were created for the best model.The nomogram visually presents the model results by formulating a scoring standard based on the regression coefficients of predictor variables. Each value level of each predictor variable is scored and then added up to obtain the total score, which is used to calculate the probability of each patient's clinical outcome. Calibration curves were employed to assess the prediction consistency of the model, while decision curve analysis (DCA) was utilized to evaluate its clinical application value.

A total of 193 patients, with an average age of 62.38 ± 10.60, were included in this study. Among them, 85 were male (44.04%), 123 had low Ki-67 expression (63.73%), and 70 had high Ki-67 expression (36.27%). Additionally, 85 cases (44.04%) were classified as having low malignant potential, while 108 cases (55.06%) were categorized as having high malignant potential. The patients were divided into a training group (154 cases) and a validation group (39 cases) at an 8:2 ratio. Statistical analysis revealed no significant difference between the two groups (P > 0.05). The median tumor volume and interquartile range were 40.23 (9.47, 123.54), with a median and interquartile range of diameter at 4.83 (3.05, 7.61). There were 117 (60.62%) cases of gastric GIST and 76 (39.38%) cases of non-gastric GIST, such as those in the small intestine and colon. In terms of morphology, 87 (45.08%) cases had regular outlines, while 106 (54.92%) had irregular outlines. Moreover, 66 (34.20%) cases were of the intraluminal type, 87 (45.08%) were of the extraluminal type, and 40 (20.73%) were classified as mixed type. Regarding tumor enhancement, 80 (41.45%) cases were uniformly enhanced, and 113 (58.55%) were heterogeneously enhanced. Additionally, 34 (17.62%) cases presented with surface ulcers, 34 (17.62%) with calcification, 73 (37.82%) with necrosis or cystic change, and 78 (40.41%) with a rich blood supply, 65 cases (33.68%) can see enlarged vessels feeding or draining the mass.The cut-off values for NLR in predicting Ki-67 expression and malignant potential are both 3.651, while the cut-off values for MLR are both 0.248. Additionally, the cut-off values for PLR in predicting Ki-67 expression and malignant potential are 279.989 and 161.660 respectively, the cut-off values for OPNI in predicting Ki-67 expression and malignant potential are 47.875 and 50.425 respectively.The detailed results can be found in Table 1.

Table 1

Baseline characteristics of the study population
Variables	Total (n = 193)	Training (n = 154)	Validation (n = 39)	t/Z/X²	P*
Age, Mean ± SD	62.38 ± 10.60	62.26 ± 11.18	62.85 ± 7.98	t=-0.31	0.758
Sex n(%)				χ²=1.91	0.167
Male	85 (44.04)	64 (41.56)	21 (53.85)
Female	108 (55.96)	90 (58.44)	18 (46.15)
Volume (cm), M (IQR)	40.23 (9.47, 123.54)	33.90 (8.69, 110.02)	49.77 (9.89, 224.55)	Z=-0.88	0.378
Diameter (cm), M (IQR)	4.83 (3.05, 7.61)	4.81 (3.01, 7.29)	5.24 (3.11, 9.29)	Z=-0.57	0.569
Ki-67,n(%)				χ²=0.18	0.669
>10%	123 (63.73)	97 (62.99)	26 (66.67)
≤ 10%	70 (36.27)	57 (37.01)	13 (33.33)
Risk classification,n(%)				χ²=0.00	0.949
Low malignant potential	85 (44.04)	68 (44.16)	17 (43.59)
High malignant potential	108 (55.96)	86 (55.84)	22 (56.41)
Tumor site,n(%)				χ²=1.79	0.181
stomach	117 (60.62)	97 (62.99)	20 (51.28)
Non stomach	76 (39.38)	57 (37.01)	19 (48.72)
Contour,n(%)				χ²=0.02	0.880
Regular	87 (45.08)	69 (44.81)	18 (46.15)
Irregular	106 (54.92)	85 (55.19)	21 (53.85)
Growth pattern,n(%)				χ²=1.10	0.578
Endophytic	66 (34.20)	55 (35.71)	11 (28.21)
Exophytic	87 (45.08)	69 (44.81)	18 (46.15)
Mixed	40 (20.73)	30 (19.48)	10 (25.64)
Enhancement pattern,n(%)				χ²=1.06	0.302
Homogeneous	80 (41.45)	61 (39.61)	19 (48.72)
Heterogeneous	113 (58.55)	93 (60.39)	20 (51.28)
Ulceration,n(%)				χ²=0.00	0.951
Absence	159 (82.38)	127 (82.47)	32 (82.05)
Presence	34 (17.62)	27 (17.53)	7 (17.95)
Calcification,n(%)				χ²=0.00	0.951
Absence	159 (82.38)	127 (82.47)	32 (82.05)
Presence	34 (17.62)	27 (17.53)	7 (17.95)
Necrosis,n(%)				χ²=0.69	0.406
Absence	120 (62.18)	98 (63.64)	22 (56.41)
Presence	73 (37.82)	56 (36.36)	17 (43.59)
IBSC,n(%)				χ²=0.41	0.520
Poor blood	115 (59.59)	90 (58.44)	25 (64.10)
Rich blood	78 (40.41)	64 (41.56)	14 (35.90)
EVFDM,n(%)				χ²=1.41	0.234
Absence	128 (66.32)	99 (64.29)	29 (74.36)
Presence	65 (33.68)	55 (35.71)	10 (25.64)
NLR,n(%)				χ²=0.70	0.404
<3.651	139 (72.02)	113 (73.38)	26 (66.67)
≥ 3.651	54 (27.98)	41 (26.62)	13 (33.33)
MLR,n(%)				χ²=1.06	0.304
<0.248	68 (35.23)	57 (37.01)	11 (28.21)
≥ 0.248	125 (64.77)	97 (62.99)	28 (71.79)
PLR, n(%)				χ²=0.12	0.724
<279.989	169 (87.56)	136 (88.31)	33 (84.62)
≥ 279.989	24 (12.44)	18 (11.69)	6 (15.38)
PLR, n(%)				χ²=1.08	0.298
<161.660	168 (87.05)	136 (88.31)	32 (82.05)
≥ 161.660	25 (12.95)	18 (11.69)	7 (17.95)
OPNI,n(%)				χ²=1.50	0.221
≥ 47.875	68 (35.23)	51 (33.12)	17 (43.59)
<47.875	125 (64.77)	103 (66.88)	22 (56.41)
OPNI, n(%)				χ²=0.47	0.491
≥ 50.425	51 (26.42)	39 (25.32)	12 (30.77)
<50.425	142 (73.58)	115 (74.68)	27 (69.23)

The statistical results of findings with p < 0.05 were showed in bold italics.

Feature selection and establishment of the predictive models of Ki-67 expression

In univariate logistic regression, factors other than calcification,age and sex showed statistically significant correlation with Ki-67 expression. Tumor volume, diameter, location, contour, enhancement pattern, EVFDM, OPNI < 47.875, and PLR ≥ 279.989 are all identified as independent predictors of GIST Ki-67 expression (P < 0.001).The comparison of regression model parameters can be observed in Fig. 2 and Table 2. The stepwise regression model emerged as the most effective for predicting Ki-67 expression, supported by its small AIC and BIC values, high HL test P value, and the largest AUC (AUC = 0.865). This model included 6 factors: diameter, growth pattern, enhancement pattern, peritumoral blood vessels, PLR ≥ 279.989, and OPNI < 47.875.

Table 2

Univariate logistic regression, multivariate logistic regression, stepwise regression, Lasso regression analysis of the risk factors for Ki-67high expression in training group, and the comparison of model parameters.
Variables	Univariate		Lasso		Stepwise		Multivariate
Variables	OR (95%CI)	P*	OR (95%CI)	P*	OR (95%CI)	P*	OR (95%CI)	P*
Age	1.01 (0.98 ~ 1.04)	0.427
Sex		0.916
Male
Female	0.96 (0.50 ~ 1.87)
Volume (cm), M (IQR)	1.01 (1.01–1.01)	< .001
Diameter (cm), M (IQR)	1.47 (1.28–1.68)	< .001	1.23 (1.01–1.50)	0.048	1.22 (1.01–1.46)	0.037	1.37(1.18–1.58)	< .001
Tumor site,n(%)
stomach
Non stomach	3.26 (1.64–6.48)	< .001	1.83 (0.75–4.43)	0.183
Contour,n(%)
Regular
Irregular	0.15 (0.07–0.33)	< .001	1.09 (0.36–3.31)	0.880
Growth pattern,n(%)
Endophytic
Exophytic	2.33 (0.97–5.60)	0.058	3.93 (1.23–12.57)	0.021	3.54 (1.14–10.94)	0.028
Mixed	0.43 (0.19–0.97)	0.041	1.24 (0.41–3.68)	0.705	1.14 (0.39–3.34)	0.806
Enhancement pattern,n(%)
Homogeneous
Heterogeneous	0.11 (0.05–0.27)	< .001	0.38 (0.12–1.17)	0.093	0.39 (0.12–1.20)	0.099
Ulceration,n(%)
Absence
Presence	3.05 (1.30–7.16)	0.010
Calcification,n(%)
Absence
Presence	0.83 (0.35–1.93)	0.659
Necrosis,n(%)
Absence
Presence	0.33 (0.17–0.66)	0.002
IBSC,n(%)
Poor blood
Rich blood	2.93 (1.49–5.76)	0.002
EVFDM,n(%)
Absence
Presence	0.16 (0.08–0.34)	< .001	0.46 (0.18–1.20)	0.112	0.43 (0.17–1.07)	0.069	2.76(1.21–6.31)	0.016
NLR
<3.651
≥ 3.651	2.97 (1.42–6.20)	0.004
MLR
<0.248
≥ 0.248	2.45 (1.19–5.04)	0.015
PLR
<279.989
≥ 279.989	7.57 (2.35–24.35)	< .001	2.97 (0.77–11.47)	0.113	3.06 (0.81–11.59)	0.099
OPNI
≥ 47.875
<47.875	4.20 (2.03–8.66)	< .001	2.15 (0.84–5.53)	0.112	2.38 (0.97–5.84)	0.058	3.31(1.44–7.65)	0.005
HL			0.362		0.590		0.745
AIC			155.587		153.360		157.418
BIC			182.920		174.619		169.567
AUC			0.864		0.865		0.832

Feature selection and establishment of the predictive models of malignant potential

In univariate logistic regression, all indicators except for calcification,age and sex showed statistically significant associations with malignant potential. Tumor volume, diameter, location, contour, enhancement pattern, growth pattern, necrosis, IBSC, EVFDM, and OPNI < 50.425 are all independent predictors of the malignant potential of GIST (P < 0.001).The comparison of model parameters can be observed in Fig. 3, and Table 3. The stepwise regression model emerged as the most effective predictive model for the malignant potential of GIST, based on its lower AIC, BIC, HL test results, and the highest AUC value (AUC = 0.950). The model included five factors: tumor volume, contour, ulcer presence, intratumoral blood supply, and OPNI < 50.425.

Table 3

Univariate logistic regression, multivariate logistic regression, stepwise regression, Lasso regression analysis of the risk factors for high malignancy potential GIST in development group, and the comparison of model parameters.
Variables	Univariate		Lasso		Stepwise		Multivariate
Variables	OR (95%CI)	P*	OR (95%CI)	P*	OR (95%CI)	P*	OR (95%CI)	P*
Age	1.01 (0.98 ~ 1.04)	0.489
Sex		0.284
Male
Female	0.70 (0.37 ~ 1.34)
Volume (cm), M (IQR)	1.07 (1.04–1.09)	< .001			1.06 (1.03–1.09)	< .001	1.06(1.03–1.09)	< .001
Diameter (cm), M (IQR)	3.10 (2.13–4.53)	< .001	2.22 (1.41–3.49)	< .001
Tumor site,n(%)
stomach
Non stomach	3.86 (1.87–7.96)	< .001
Contour,n(%)
Regular
Irregular	0.06 (0.02–0.12)	< .001	0.57 (0.17–1.89)	0.361	0.17 (0.05–0.62)	0.066
Growth pattern,n(%)
Endophytic
Exophytic	0.95 (0.37–2.43)	0.919
Mixed	0.17 (0.08–0.37)	< .001
Enhancement pattern,n(%)
Homogeneous
Heterogeneous	0.06 (0.03–0.14)	< .001	0.53 (0.16–1.74)	0.294
Ulceration,n(%)
Absence
Presence	2.64 (1.04–6.68)	0.040			0.16 (0.03–0.98)	0.094
Calcification,n(%)
Absence
Presence	0.47 (0.19–1.15)	0.099
Necrosis,n(%)
Absence
Presence	0.15 (0.07–0.33)	< .001
IBSC,n(%)
Poor blood
Rich blood	7.14 (3.34–15.24)	< .001	3.50 (1.13–10.78)	0.029	5.27 (1.57–17.69)	0.008	4.62(1.53–13.92)	0.007
EVFDM,n(%)
Absence
Presence	0.07 (0.03–0.19)	< .001	0.50 (0.14–1.72)	0.271
NLR
<3.651
≥ 3.651	3.27 (1.47–7.29)	0.004
MLR
<0.248
≥ 0.248	2.44 (1.25–4.76)	0.009
PLR
<161.660
≥ 161.660	2.29 (1.20–4.39)	0.012
OPNI
≥ 50.425
<50.425	0.15 (0.06–0.34)	< .001	0.35 (0.09–1.37)	0.132	0.22 (0.05–0.96)	0.045	5.01(1.22–20.54)	0.025
HL			0.810		0.682		0.900
AIC			100.134		96.33		96.864
BIC			123.393		114.552		109.012
AUC			0.947		0.950		0.943

Predictive Performance of two models

The Ki-67 expression prediction model showed an AUC of 0.865 (95% CI: 0.807–0.922) in the training group with a sensitivity of 0.835 and specificity of 0.772. The positive predictive value (PPV) and negative predictive value (NPV) were 0.862 and 0.733, respectively, with a cut-off value of 0.437. In the validation group, the AUC was 0.784 (95% CI: 0.631–0.937) with a sensitivity of 0.731 and specificity of 0.615.The PPV and NPV were 0.792 and 0.533, respectively, with the same cut-off value of 0.437. For the malignant potential prediction model, the training group had an AUC of 0.950 (95% CI: 0.920–0.980) with a sensitivity of 0.956 and specificity of 0.837. The positive predictive value (PPV) and negative predictive value (NPV) were 0.823 and 0.960, respectively, with a cut-off value of 0.570. In the validation group, the AUC was 0.936 (95% CI: 0.867-1.000) with a sensitivity of 0.765 and specificity of 0.864. The PPV and NPV were 0.812 and 0.826, respectively, with the same cut-off value of 0.570.

Table 4

Table shows the performances of two models for the training and validation populations.
	Ki-67 expression model		Malignancy potential model
	Training	Validation	Training	Validation
AUC (95%CI)	0.865(0.807–0.922)	0.784 (0.631–0.937)	0.950(0.920–0.980)	0.936(0.867-1.000)
Accuracy(95%CI)	0.812(0.741–0.870)	0.692(0.524–0.830)	0.890(0.829–0.934)	0.821(0.665–0.925)
Sensitivity(95%CI)	0.835 (0.761–0.909)	0.731(0.560–0.901)	0.956 (0.907–1.000)	0.765(0.563–0.966)
Specificity(95%CI)	0.772 (0.663–0.881)	0.615(0.351–0.880)	0.837(0.759–0.915)	0.864 (0.720-1.000)
PPV (95%CI)	0.862 (0.792–0.931)	0.792(0.629–0.954)	0.823(0.739–0.907)	0.812(0.621- 1.000)
NPV (95%CI)	0.733 (0.621–0.845)	0.533(0.281–0.786)	0.960 (0.916–1.000)	0.826(0.671–0.981)
cut off	0.437	0.437	0.570	0.570

The nomogram, calibration curve and clinical decision curve of two models

Based on the nomogram results, each variable's score aligns with the score on the top axis, and when summed, these scores form the total score. The corresponding value on the bottom axis respectively indicates the high Ki-67 expression and high malignant potential risk probability of GIST. When the total score of Ki-67 exceeds the cut-off value of 134.351, it signifies high Ki-67 expression. Similarly, surpassing the cut-off value of 14.359 for malignant potential indicates a high level of malignancy. Figure 4 and Fig. 5 present the nomogram, calibration curve, and clinical decision curve of the Ki-67 prediction model, while Fig. 6 and Fig. 7 display the same for the malignant potential model. The calibration curves and clinical decision curves of both models demonstrate strong agreement with observed results, affirming the reliability and practicality of these prediction models in anticipating high Ki-67 expression and malignant potential in GIST cases prior to surgery.

This study utilized three regression methods to identify better risk factors associated with high Ki-67 expression and malignant potential. The optimal fusion model was then established and chosen based on the findings. Subsequently, a nomogram, calibration curve, and clinical decision curve were developed for the selected prediction model. The results indicate that both models exhibit strong discriminatory ability and practical utility. Imaging plays a crucial role in the diagnosis, evaluation, and treatment of GIST. When GIST is identifiable on imaging and can be completely excised surgically, preoperative biopsy is deemed unnecessary ^[17] .To enhance non-invasive preoperative assessment, numerous scholars, both domestic and foreign, have developed prediction models for Ki-67 expression or malignant potential. Commonly utilized imaging features include tumor diameter, growth pattern, enhancement mode, contour, necrosis, and peritumoral blood vessels^༻4–7༽. This study yielded similar results.In order to further improve the prediction performance, this study compared the correlation of tumor volume and diameter, IBSC and EVFDM, OPNI and NLR, MLR, PLR with Ki-67 expression and malignant potential in multivariate regression analysis. By incorporating better indicators into the model, higher prediction performance was achieved.

The size of the tumor is a crucial factor in determining the grading of GIST ^[18,19]. While the diameter measurement method is commonly used in current studies due to its simplicity, research suggests that tumor volume is a more significant predictor of malignancy. Tumor volume, which takes into account the three-dimensional nature of tumors, provides a more accurate reflection of tumor size compared to diameter alone. Given the irregular shape of some tumors, the layer-by-layer delineation method is considered more reliable for calculating volume than the ellipsoid volume formula ^[20,21].But the results indicate that diameter may be a more accurate measure of GIST proliferation ability compared to volume. This could be due to the irregular shape and unidirectional growth of lesions with higher proliferation rates, making diameter a more closely correlated factor. There is currently some debate regarding the use of a 5cm boundary for measuring tumor diameter in both domestic and foreign literature^[22,23]. Discrepancies between imaging measurements and postoperative pathology have been noted. Enhanced CT and endoscopic ultrasound techniques have been found to underestimate GIST size to some extent^[24,25]. Some studies have also suggested that the NIH classification may not be optimal for determining tumor size and mitosis thresholds^[18]. Therefore, in this study, image-measured diameter was used in statistical analysis to minimize errors and allow for individualized evaluation, rather than relying on a 5cm cutoff. The experiment revealed that this approach resulted in higher prediction performance (AUC) and enabled individualized assessment.

Tumor blood vessels, which include both intratumoral and peritumoral blood vessels, have been widely recognized as crucial for cancer progression. They supply tumors with essential oxygen and nutrients for growth, as well as serve as pathways for tumor metastasis through blood or lymphatic vessels ^[4]. The presence of peritumoral blood vessels is a reliable indicator of malignancy in GIST, but intratumoral blood vessels also contribute significantly to tumor development. In comparison to peritumoral blood vessels, intratumoral blood vessels serve as a more reliable indicator of the malignant potential of GIST. The identification of tumor invasiveness across different grades primarily hinges on the development of new blood vessels^[26]. While ultrasound research suggests that GISTs exhibiting regular blood flow from the periphery to the center are more likely to possess high malignant potential^༻27༽, studies on CT imaging remain scarce. CT scans have the capability to detect subcentimeter high-vascular lesions^[17], so this study conducted a preliminary discussion on the subject, referencing the above standards. Additionally, it found statistical significance between IBSC and Ki-67 expression (P = 0.002). High Ki-67 expression is linked to increased microvessel density and tumor neovascularization^༻5༽. The form and density of blood vessels within a tumor are closely linked to its growth, diversity, enhancement patterns, and level of enhancement ^[28]. The assessment of tumor invasiveness across various grades is largely determined by the development of new blood vessels ^[26]. The formation and evolution of microvessels within tumors are commonly utilized in the prognosis and evaluation of treatment effectiveness for numerous types of cancer ^[29].Interestingly, EVFDM appear to be more predictive of Ki-67 expression compared to IBSC. This could be due to inadequate tumor blood supply leading to necrosis within the tumor, hindering new blood vessel observation. Furthermore, the relationship between IBSC and EVFDM is not linear, suggesting a need for further investigation into the underlying mechanisms.

Inflammation has been linked to different stages of tumor development ^[30]. Preoperative serological indicators can provide valuable supplementary evidence for assessing tumors. While NLR, MLR, and PLR are commonly used in predicting and evaluating various tumors ^[31], there have been no studies utilizing OPNI to develop a GIST prediction model. Low OPNI is considered an independent prognostic indicator for GIST patients and is superior to NLR and PLR ^[32], highlighting the potential of OPNI in predicting Ki-67 expression and malignant potential. Results from this study demonstrate that, compared to other serological indicators, OPNI is a more accurate predictor of GIST Ki-67 and malignant potential. This is because OPNI incorporates nutritional and immune factors; hypoalbuminemia and low lymphocyte levels create a conducive environment for tumor growth and progression ^[33]. In a recent 2024 study, MLR combined with imaging features was used to predict the malignant potential of gastrointestinal GIST, achieving an AUC of 0.935 ^[34]. This study discovered that only three indicators, namely tumor volume, IBSC, and OPNI, were statistically significant in predicting the malignant potential of gastrointestinal GIST (P < 0.05) using multi-factor logistic regression analysis. The prediction model constructed had an AUC of 0.943, further supporting the notion that volume,IBSC, and OPNI are superior indicators for predicting the malignant potential of GIST.In the stepwise regression analysis, contour and ulcer were added to the model construction along with volume, IBSC, and OPNI. The resulting model achieved an AUC of 0.950.

This study has several limitations. Firstly, it is a single-center study and lacks multi-center and prospective studies for validation. Secondly, the underlying mechanism between Ki-67 and serological indicators requires further investigation by biologists. Additionally, there is currently no standardized method for enhanced CT in determining intratumoral blood supply conditions. This study serves as a preliminary assessment, awaiting comparison with pathological results and advancements in CT technology. Lastly, the correlation between intratumoral new blood vessels and peritumoral blood vessels warrants further exploration.

This study aims to provide valuable information to clinicians, enabling them to conduct a more thorough and systematic non-invasive assessment of GIST before surgery. The models utilizes easily accessible imaging characteristics and serological indicators, making it clinically practical. By utilizing these two models, clinicians can make more informed treatment decisions, ultimately leading to improved patient outcomes. Otherwise, the Ki-67 expression model may have the potential to enhance the assessment of small GISTs and early-stage GISTs with a high risk of malignancy.

The maximum diameter, growth pattern, EVFDM, peritumoral vessels, PLR, OPNI are correlated with Ki-67 high expression. Volume, contour, ulcer, IBSC and OPNI are correlated with malignant potential of gastrointestinal stromal tumors; Preoperative models developed using enhanced CT imaging can respectively predict the malignant potential and Ki-67 expression of GIST. Furthermore, when combined with serological indicators, the predictive accuracy of this model can be further enhanced.

GIST Gastrointestinal stromal tumor
CE-CT contrast-enchanced computed tomography

ROC Receiver operating characteristic
AUC Area under the curve

C.I. Confidence interval
EVFDM enlarged vessels feeding or draining the mass

IBSC intratumoral blood supply conditions

Acknowledgements We thank all patients, their families and all investigators involved in the present study. I would also like to thank my tutor, Professor Weizhi Chen.

Author Contributions All authors contributed to the study conception and design. Material preparation, data collection and analysis were performed by Jun Tian and Weizhi Chen. The first draft of the manuscript was written by Jun Tian, and all authors commented on previous versions of the manuscript. All authors read and approved the final manuscript.

Funding No funding

Conflict of interest The authors declare no conflict of interest.

Data statementThe authors declare that they had full access to all of the data in this study and the authors take complete responsibility for the integrity of the data and the accuracy of the data analysis.

Institutional review board statement The study was conducted according to the guidelines of the Declaration of Helsinki and approved by the institutional review board.

Informed consent This study is a retrospective study, and patients' informed consent was waived with the approval of the institutional review board. This study was approved by the Ethics Review Committee of the First Affiliated Hospital of Jinzhou Medical University, with ethics number KYLL202445.

Consent for publication Not applicable.

Author details The First Affiliated Hospital of Jinzhou Medical University Declarations.

Disclosures Jun Tian and Weizhi Chen have no conflicts of interest or financial ties to disclose.

Blay, J. Y., Kang, Y. K., Nishida, T., & von Mehren, M. (2021). Gastrointestinal stromal tumours. Nature reviews. Disease primers, 7(1), 22.
Li, G. Z., & Raut, C. P. (2019). Targeted therapy and personalized medicine in gastrointestinal stromal tumors: drug resistance, mechanisms, and treatment strategies. OncoTargets and therapy, 12, 5123–5133.
Zhao Yan, Wang Yuhao, Wang Juan&Yang Jianjun (2024). Diagnosis and treatment progress of gastrointestinal stromal tumors. Chinese Journal of General Surgery (electronic version) (01), 66-70
Wang, Y., Bai, G., Zhang, H., & Chen, W. (2023). Simple Scoring Model Based on Enhanced CT in Preoperative Prediction of Biological Risk of Gastrointestinal Stromal Tumor. Technology in cancer research & treatment, 22, 15330338231194502.
Yuan Shichao, Zhao Qing&Fan Hongyan (2021). The predictive value of dual energy CT for Ki-67 expression in gastric stromal tumors and its relationship with pathological parameters. Clinical meta-analysis (11), 1013-1018
Ren, C., Wang, S., & Zhang, S. (2020). Development and validation of a nomogram based on CT images and 3D texture analysis for preoperative prediction of the malignant potential in gastrointestinal stromal tumors. Cancer imaging : the official publication of the International Cancer Imaging Society, 20(1), 5.
Liu, Y., He, C., Fang, W., Peng, L., Shi, F., Xia, Y., Zhou, Q., Zhang, R., & Li, C. (2023). Prediction of Ki-67 expression in gastrointestinal stromal tumors using radiomics of plain and multiphase contrast-enhanced CT. European radiology, 33(11), 7609–7617.
Zhang, Q. W., Gao, Y. J., Zhang, R. Y., Zhou, X. X., Chen, S. L., Zhang, Y., Liu, Q., Xu, J. R., & Ge, Z. Z. (2020). Personalized CT-based radiomics nomogram preoperative predicting Ki-67 expression in gastrointestinal stromal tumors: a multicenter development and validation cohort. Clinical and translational medicine, 9(1), 12.
Yang, Z., Feng, X., Zhang, P., Chen, T., Qiu, H., Zhou, Z., Li, G., Tao, K. X., Li, Y., & China Gastrointestinal Stromal Tumor Study Group (CN-GIST) (2019). Clinicopathological features and prognosis of 276 cases of primary small (≤ 2 cm) gastric gastrointestinal stromal tumors: a multicenter data review. Surgical endoscopy, 33(9), 2982–2990.
Wang, J. P., Liu, L., Li, Z. A., Wang, Q., Wang, X. Y., & Lin, J. (2021). Ki-67 labelling index is related to the risk classification and prognosis of gastrointestinal stromal tumours: a retrospective study. Gastroenterologia y hepatologia, 44(2), 103–114.
Zhu, M. P., Ding, Q. L., Xu, J. X., Jiang, C. Y., Wang, J., Wang, C., & Yu, R. S. (2022). Building contrast-enhanced CT-based models for preoperatively predicting malignant potential and Ki67 expression of small intestine gastrointestinal stromal tumors (GISTs). Abdominal radiology (New York), 47(9), 3161–3173.
Huang Liebin, Long Wansheng, Chen Qinxian, Xue Huimin, Huang Wensi, Zhou Tao& Li Qing (2021). Research on the construction of a postoperative recurrence prediction model for gastric stromal tumors based on CT and pathology. Radiological Practice (06), 762-766.
Li Xuhua, Chen Yulan&Lai Fang (2023). The relationship between Ki-67 marker index and clinical pathological characteristics and prognosis of primary resectable gastrointestinal stromal tumors. Medical Theory and Practice (15), 2625-2627.
Wei Xiaohua, Kong Ying, Liu Hong, Jiang Xuehui, Kaisaner Rexiti&Chen Xu (2024). Chinese expert consensus on individualized medication management of imatinib, a targeted drug for gastrointestinal stromal tumors. Chinese Pharmacy (03), 257-270
Kim, G. H., Ahn, J. Y., Gong, C. S., Kim, M., Na, H. K., Lee, J. H., Jung, K. W., Kim, D. H., Choi, K. D., Song, H. J., Lee, G. H., & Jung, H. Y. (2020). Efficacy of Endoscopic Ultrasound-Guided Fine-Needle Biopsy in Gastric Subepithelial Tumors Located in the Cardia. Digestive diseases and sciences, 65(2), 583–590.
Jr，F.E.H.rms:Regression Modeling Strategies.(2023).
Serrano, C., Martín-Broto, J., Asencio-Pascual, J. M., López-Guerrero, J. A., Rubió-Casadevall, J., Bagué, S., García-Del-Muro, X., Fernández-Hernández, J. Á., Herrero, L., López-Pousa, A., Poveda, A., & Martínez-Marín, V. (2023). 2023 GEIS Guidelines for gastrointestinal stromal tumors. Therapeutic advances in medical oncology, 15, 17588359231192388.
Trinh, V. Q., Dashti, N. K., & Cates, J. M. M. (2021). A proposed risk assessment score for gastrointestinal stromal tumors based on evaluation of 19,030 cases from the National Cancer Database. Journal of gastroenterology, 56(11), 964–975.
Wang Ming&Cao Hui (2021). Interpretation of Precision Diagnosis and Treatment of Gastrointestinal Stromal Tumors from Major Guidelines Update Changes at Home and Abroad in 2020. Chinese Journal of Practical Surgery (02), 125-129.
Lino-Silva, L. S., Segales-Rojas, P., Aguilar-Cruz, E., Salcedo-Hernández, R. A., & Zepeda-Najar, C. (2019). Gastrointestinal Stromal Tumors Risk of Recurrence Stratification by Tumor Volume is a Best Predictor Compared with Risk Based on Mitosis and Tumor Size. Journal of gastrointestinal cancer, 50(3), 513–518.
Mazzei, M. A., Cioffi Squitieri, N., Vindigni, C., Guerrini, S., Gentili, F., Sadotti, G., Mercuri, P., Righi, L., Lucii, G., Mazzei, F. G., Marrelli, D., & Volterrani, L. (2020). Gastrointestinal stromal tumors (GIST): a proposal of a "CT-based predictive model of Miettinen index" in predicting the risk of malignancy. Abdominal radiology (New York), 45(10), 2989–2996.
Li, C., Fu, W., Huang, L., Chen, Y., Xiang, P., Guan, J., & Sun, C. (2021). A CT-based nomogram for predicting the malignant potential of primary gastric gastrointestinal stromal tumors preoperatively. Abdominal radiology (New York), 46(7), 3075–3085.
Cannella, R., Tabone, E., Porrello, G., Cappello, G., Gozzo, C., Incorvaia, L., Grignani, G., Merlini, A., D'Ambrosio, L., Badalamenti, G., Regge, D., & Bartolotta, T. V. (2021). Assessment of morphological CT imaging features for the prediction of risk stratification, mutations, and prognosis of gastrointestinal stromal tumors. European radiology, 31(11), 8554–8564.
Li, Y., Chen, X., Ma, X., & Lu, X. (2022). Computed tomography in the size measurement of gastric gastrointestinal stromal tumors: Implication to risk stratification and "wait-and-see" tactics. European journal of surgical oncology : the journal of the European Society of Surgical Oncology and the British Association of Surgical Oncology, 48(8), 1739–1745.
Apte, S. S., Radonjic, A., Wong, B., Dingley, B., Boulva, K., Chatterjee, A., Purgina, B., Ramsay, T., & Nessim, C. (2021). Preoperative imaging of gastric GISTs underestimates pathologic tumor size: A retrospective, single institution analysis. Journal of surgical oncology, 124(1), 49–58.
Feng, C., Lu, F., Shen, Y., Li, A., Yu, H., Tang, H., Li, Z., & Hu, D. (2018). Tumor heterogeneity in gastrointestinal stromal tumors of the small bowel: volumetric CT texture analysis as a potential biomarker for risk stratification. Cancer imaging : the official publication of the International Cancer Imaging Society, 18(1), 46.
Yamashita, Y., Kato, J., Ueda, K., Nakamura, Y., Abe, H., Tamura, T., Itonaga, M., Yoshida, T., Maeda, H., Moribata, K., Niwa, T., Maekita, T., Iguchi, M., Tamai, H., & Ichinose, M. (2015). Contrast-enhanced endoscopic ultrasonography can predict a higher malignant potential of gastrointestinal stromal tumors by visualizing large newly formed vessels. Journal of clinical ultrasound : JCU, 43(2), 89–97.
Chen, W. T., Huang, C. J., Wu, M. T., Yang, S. F., Su, Y. C., & Chai, C. Y. (2005). Hypoxia-inducible factor-1alpha is associated with risk of aggressive behavior and tumor angiogenesis in gastrointestinal stromal tumor. Japanese journal of clinical oncology, 35(4), 207–213.
Braman, N., Prasanna, P., Bera, K., Alilou, M., Khorrami, M., Leo, P., Etesami, M., Vulchi, M., Turk, P., Gupta, A., Jain, P., Fu, P., Pennell, N., Velcheti, V., Abraham, J., Plecha, D., & Madabhushi, A. (2022). Novel Radiomic Measurements of Tumor-Associated Vasculature Morphology on Clinical Imaging as a Biomarker of Treatment Response in Multiple Cancers. Clinical cancer research : an official journal of the American Association for Cancer Research, 28(20), 4410–4424.
Mantovani, A., Allavena, P., Sica, A., & Balkwill, F. (2008). Cancer-related inflammation. Nature, 454(7203), 436–444.
Yapar, A., Tokgöz, M. A., Yapar, D., Atalay, İ. B., Ulucaköy, C., & Güngör, B. Ş. (2021). Diagnostic and prognostic role of neutrophil/lymphocyte ratio, platelet/lymphocyte ratio, and lymphocyte/monocyte ratio in patients with osteosarcoma. Joint diseases and related surgery, 32(2), 489–496.
Wang, F., Tao, T., Yu, H., Xu, Y., Yang, Z., Xia, X., Wang, M., Zong, L., & Guan, W. (2021). Prognostic value of Onodera's nutritional index for intermediate- and high-risk gastrointestinal stromal tumors treated with or without tyrosine kinase inhibitors. World journal of surgical oncology, 19(1), 227.
Song, D. J., & Yang, L. T. (2022). Zhonghua wei chang wai ke za zhi = Chinese journal of gastrointestinal surgery, 25(12), 1138–1143.
Liu, Z., Gao, J., Zeng, C., & Chen, Y. (2024). Development and validation of a preoperative risk nomogram prediction model for gastric gastrointestinal stromal tumors. Surgical endoscopy, 38(4), 1933–1943.

No competing interests reported.

Prediction of Ki-67 expression and malignant potential in gastrointestinal stromal tumors: novel models based on CE-CT and serological indicators

Status:

Version 1

Abstract

Figures

Introduction

Materials and Methods

Patients

Histopathologic examination

Serological indicators

CT imaging acquisition

Image analysis

Statistical analysis

Result

Feature selection and establishment of the predictive models of Ki-67 expression

Feature selection and establishment of the predictive models of malignant potential

Predictive Performance of two models

The nomogram, calibration curve and clinical decision curve of two models

Discussion

Conclusion

Abbreviations

Declarations

References

Additional Declarations

Status:

Version 1