Prognostic Predicting Model of Pancreatic Body Tail Carcinoma Using Clinical and CT Radiomic Data

Introduction

In recent years, with the rapid growth of China's industrial economy, issues like environmental pollution, stress, and eating disorder are getting more and more serious, and the incidence of pancreatic tumors is surging in China, reaching a number of 5.1/100 000. At the same time, pancreatic cancer is challenging to treat and has a poor prognosis with a median survival of 6 months reported by nature journal.^1–4 Pancreatic body tail cancer (PBTC) accounted for 20% of pancreatic cancers. Because of its occult location as well as the lack of specific clinical presentations and tumor markers in the early stage, 70% of the PBTC patients were in the middle and late stages when diagnosed. In addition, the complete surgical resection rate was less than 40% due to commonly coeliac trunk and superior mesenteric artery involvement or liver-intestinal metastasis.^5–7 Although a small portion of PBTC patients gained the opportunity of surgical resection, they often suffered from uncontrollable recurrence and death because of the high vascular and nerve invasion, lymph node metastasis characteristics, and poor sensitivity to current chemoradiotherapy and immunotherapy. The 5-year overall survival was less than 10%. Therefore, early preoperative diagnosis and outcome prediction of PBTC are crucial to adjust the treatment plan to prolong the survival period as much as possible.^8,9 More specifically, outcome prediction of PBTC may effect its treatment, The results of this study indicate that for older PBTC patients with diabetes and peripancreatic lymph node swelling, intensive care management, dietary treatment or insulin therapy for blood glucose control, and minimally invasive clearance or proton heavy ion therapy of peripheral lymph nodes can effectively reduce the incidence of postoperative complications; PBTC with carcinoembryonic antigen (CEA) increased cases should be vigilant in getting a short-term ultrasound or MR rechecks to detect surrounding tissue or liver metastasis early and undergo the targeted treatment; for PBTC cases with more peripancreatic fluid, dilation of the pancreatic ducts and bile ducts, we recommend that patients undergo 1 to 3 days abdominal external drainage, percutaneous transhepatic cholangial drainage, and anti-inflammatory treatment before operation. In addition, for PBTC with poor prognosis prediction, longer regular targeted chemotherapy, molecular targeted drugs (erlotinib, everolimus, or sunitinib), and regular imaging rechecks should be conducted to maximize the life expectancy of patients after surgery. The development of radiomics technology, which started in 2012, has greatly promoted the evolution of imaging diagnosis. Now it has been widely used in the artificial intelligence recognition of liver cancer, kidney cancer, and lung nodules, and has achieved better clinical benefits. In the same way, our team applied computed tomography (CT) radiomics technology to the diagnosis and treatment evaluation of PBTC and obtained good benefits. The results confirmed that the radiomics score (Radscore), surgery in the specialized medical center of pancreatic disease, nerve invasion, etc, were independent factors affecting the prognosis of PBTC; the comprehensive radiomics score obtained by CT image segmentation of pancreatic mass may represent the basic texture information of the mass image and reflects its essential characteristics (including mass heterogeneity, possible fat or cancerous soft tissue replacement, dynamic enhancement mode, tissue necrosis, tissue ischemia, enhancement mode of PTBC dense tissue, local tissue sclerosis of pancreas, etc). There are few reports on clinical data combined with radiomics to predict the prognosis of PBTC after consulting the references in PUBMED, Therefore, this study has some novelty that may be helpful to the prognosis management of PTBC. This study aimed to establish a model to predict the prognosis of PBTC based on clinicopathological-CT imaging features, and achieved good results, providing a reference for patients’ prognosis prediction and targeted treatment.

Materials and Methods

The clinical and imaging data of 203 patients who underwent surgical resection or endoscopic ultrasound biopsy and were pathologically diagnosed with PBTC between December 2012 and December 2022 in the Pancreatic Surgery Department of Xiangyang NO.1 People's Hospital were retrospectively collected. All PBTC patients had complete preoperational abdominal enhancement and chest CT examinations. The data included age, gender, diabetes condition, hypertension condition, CEA, carbohydrate antigen 19-9 (CA19-9), tumor location, tumor size, tumor, node, and metastasis (TNM) stage, CT enhancement mode, and CT texture parameters. Inclusion criteria: patients accepted surgical treatment with complete clinicopathological, imaging, and follow-up data. Exclusion criteria: patients accepted conservative treatment, or with pancreatic metastases from other primary malignancies; combined with other tumors; and combined with multiorgan dysfunction syndrome (Figure 1).^4,9 Finally, a total of 143 PBTC patients were included, and they all signed the informed consent. And then PBTC was divided into a good prognosis group (survival time ≥ 6 months) and a poor prognosis group (survival time < 6 months) according to a median survival of 6 months reported by nature journal. ⁴

OPEN IN VIEWER

Figure 1.

The simplified inclusion and exclusion criteria for patient enrollment in the present study.

This is an exploratory, single-center, and retrospective study. This study was approved by the local Institutional Review Board (Ethics number: No. 1126 [2020]), and conducted in accordance with the Declaration of Helsinki. This study obtained informed consent from participants. The reporting of this study conforms to STROBE guidelines. ¹⁰ We have de-identified patient details such that the identity of any person may not be ascertained in any way. This study has followed relevant Equator guidelines ¹¹ and started in March 2021 and ended in December 2022.

CT Image Segmentation, Texture Feature Extraction, and Selection

All CT data were imported into a 3D slicer software (version 5.2.2, https://download.slicer.org/) and read under abdominal window (width 400/40 HU) settings. The whole PBTC lesion was selected for carcinoma volume segmentation after data acquisition. A radiographer and radiologist without knowledge of the pathology results manually delineated the volumes of interest (VOI) layer by layer, respectively. The PBTC’s VOI must enclose the entire mass as much as possible, including the fat, cyst, vasculum, necrosis, etc within the mass, and exclude the normal pancreatic tissue. Inter- and intraclass correlation coefficients (ICCs) were used to calculate the inter- and intraobserver reproducibility, and reliability of texture feature extraction. In this study, ICC is required to be greater than 0.7.

Informatics-radiomics module in 3D slicer was used to extract a total of 897 × 3 radiomics features, including histogram, gray level co-occurrence matrix (GLCM), gray level run length matrix (GLRLM), gray level size zone matrix (GLSZM), neighboring gray-tone difference matrix (NGTDM), gray level dependence matrix (GLDM), etc. Lasso regression with 10-fold cross-validation was performed to extract effective texture data, and then a radiomics label was constructed, and the radiomics score (Radscore) of each CT enhanced phase was calculated accordingly after feature screening. Radscore1 was for arterial phase, Radscore2 was for venous phase, and Radscore3 was for delayed phase (Figure 2).^12–14

OPEN IN VIEWER

Figure 2.

Sketch map of texture parameter extraction of pancreatic tumor CT image using 3D slicer.

Result

Kaplan-Meier Analysis of Factors Affecting PBTC Recurrence

The univariate analysis demonstrated that Age, diabetes, TNM stage, CT enhancement mode, nerve invasion, peripancreatic lymph node swelling, advanced hospital surgery, tumor size, CA19-9, CEA, Radscore1/2/3 were the influencing factors of PBTC recurrence (P < .05), while gender, lesion morphology, dilatation of main pancreatic duct, acute obstructive pancreatitis, pancreatic atrophy, vascular invasion, BMI were not significantly correlated with the recurrence of PBTC (Table 1 and Figure 3).

OPEN IN VIEWER

Figure 3.

Kaplan-Meier analysis diagram (Log-rank method) of pancreatic body tail cancer (PBTC) recurrence based on the above risk factors.

OPEN IN VIEWER

Table 1.

Kaplan-Meier Analysis of Factors Affecting PBTC Recurrence.

Factors	N	Median survival (month)	Hazard ratio	P
Age (year)			1.02 (1.00-1.03)	.038*
＜ 48	21	7.61 (5.77-9.81)
≥ 48	122	5.53 (5.33-5.80)
Gender			1.17 (0.79-1.72)	.436
Male	79	5.72 (5.54-5.95)
Female	64	5.41 (5.05-5.90)
Diabetes condition			1.51 (1.01-2.23)	.044*
Yes	83	5.33 (5.13-5.76)
No	60	5.88 (5.64-9.52)
Hypertension condition			1.14 (0.76-1.70)	.537
Yes	96	5.65 (5.36-5.92)
No	47	5.66 (5.31-7.66)
Lesion morphology			0.887 (0.59-1.28)	.470
Swelling and protruding growth	66	5.66 (5.39-5.80)
Localized growth	77	5.64 (5.29-7.66)
TNM stage			1.51 (1.02-2.22)	.042*
I + II stage	71	5.83 (5.53-9.52)
III + IV stage	72	5.49 (5.29-5.80)
CT enhancement mode			1.53 (1.02-2.30)	.038*
Homogeneous enhancement	59	5.80 (5.59-19.18)
Slight fast enhancement	84	5.36 (5.22-5.83)
Dilation of the main pancreatic duct			1.47 (0.98-2.19)	.060
＜ 5 mm	60	5.87 (5.54-11.30)
≥ 5 mm	83	5.50 (5.16-5.77)
Acute obstructive pancreatitis			1.29 (0.86-1.93)	.219
Yes	90	5.47 (5.29-5.71)
No	53	5.92 (5.77-9.52)
Pancreatic atrophy			1.02 (0.69-1.51)	.914
Yes	70	5.59 (5.31-5.92)
No	73	5.66 (5.39-5.97)
Vascular invasion			1.14 (0.76-1.72)	.530
Yes	92	5.71 (5.36-5.90)
No	51	5.56 (5.33-11.30)
Nerve invasion			1.88 (1.20-2.95)	.006*
Yes	100	5.39 (5.28-5.74)
No	43	8.58 (5.66-13.3)
Peripancreatic lymph node swelling			1.61 (1.05-2.46)	.028*
Yes	105	5.33 (5.05-5.74)
No	38	5.77 (5.53-7.65)
Peripancreatic fluid			1.34 (0.91-1.98)	.141
Yes	62	5.39 (5.16-5.74)
No	81	5.80 (5.56-8.16)
Advanced hospital surgery			0.62 (0.42-0.90)	.012*
Yes	97	5.77 (5.54-8.58)
No	46	5.35 (5.02-5.76)
BMI (kg/m2)			0.98 (0.94-1.03)	.434
＜ 31.3	125	5.59 (5.39-5.81)
≥ 31.3	18	7.65 (5.16-12.33)
Tumor size (mm)			1.01 (1.00-1.02)	.005*
＜ 74.8	127	5.76 (5.50-5.95)
≥ 74.8	16	5.01 (3.78-5.71)
CA19-9 (kU/L)			1.01 (1.00-1.01)	.018*
＜ 113.33	69	5.90 (5.64-11.30)
≥ 113.33	74	5.32 (5.05-5.72)
Serum amylase (U/L)			1.01 (1.00-1.05)	.233
＜ 124.74	112	5.77 (5.53-5.97)
≥ 124.74	31	5.13 (4.80-5.59)
CEA (µg/L)			1.01 (1.00-1.02)	.010*
＜142.03	21	11.30 (7.66-14.22)
≥ 142.03	122	5.54 (5.34-5.77)
Radscore 1			1.53 (1.18-1.97)	.001*
＜6.45	45	9.52 (5.71-15.74)
≥ 6.45	98	5.49 (5.28-5.74)
Radscore 2			1.60 (1.38-1.86)	.001*
＜5.96	47	9.96 (8.98-13.34)
≥ 5.96	96	5.33 (5.15-5.54)
Radscore 3			1.54 (1.20-1.96)	.001*
< 6.46	50	8.43 (5.80-14.56)
≥ 6.46	93	5.39 (5.22-5.71)

Abbreviations: PBTC, pancreatic body tail cancer; TNM, tumor, node, and metastasis; CA19-9, carbohydrate antigen 19-9; CEA, carcinoembryonic antigen; CT, computed tomography; BMI, body mass index.

Establishment of 3 Predictive Models by Logistic Regression and Model Validation

According to the median survival of 6 months, the PBTC patients were divided into 2 groups, a group of good prognoses and a group of poor prognoses, followed by the establishment of a training set and a test set at a ratio of 7/3. Multivariate logistic analysis confirmed that nerve invasion, advanced hospital surgery, dilation of the main pancreatic duct, Radscore2 were independent factors affecting the death/survivals of PBTC patients (P < .05). In the training set, the combined model achieved the best predictive performance [AUC 0.976, 95% CI (0.924-0.996)], significantly higher than that of the clinicopathological model [AUC 0.821, 95% CI (0.732-0.890), P = .0001] and that of the CT radiomics model [AUC 0.943, 95% CI (0.879-0.980), P = .0623]. Also, in the test set, the combined model achieved the best predictive performance [AUC 0.944, 95% CI (0.826-0.991)], significantly superior to the clinicopathological model [AUC 0.770, 95% CI (0.615-0.886), P = .0145], and the CT radiomics model [AUC 0.883, 95% CI (0.746-0.961), P = .1311]. DCA confirmed that the clinical net benefits of the combined model were better. The nomogram simplified the process of predicting the prognosis of PBTC in clinical practice (Tables 2 to 4 and Figures 4 to 6).

OPEN IN VIEWER

Figure 4.

DeLong nonparametric method was used to estimate the area under the curve of ROC between different prediction models of the training set (a) and test set (b) and compare its effectiveness in predicting the PBTC prognoses. The area under the curve of the combined model was the largest.

OPEN IN VIEWER

Figure 5.

In the training set (a) and the test set (b), the better clinical net benefit of the combined model (C) was analyzed and confirmed using a decision curve.

OPEN IN VIEWER

Figure 6.

Based on the combined model, we created a new nomogram (A) and calibration curve (B), which simplified the process of predicting the prognosis of pancreatic body tail cancer (PBTC) in clinical practice (R2: Radscore 2).

OPEN IN VIEWER

Table 2.

Logistic Regression Analysis Results of Clinicopathological Model Based on Clinical/Pathological Characteristics for Predicting PBTC Prognoses.

Clinicopathological model	Univariate analysis		Multivariate analysis
Factors	P	Hazard ratio	P	Hazard ratio
Age	.022*	1.036 (1.005-1.070)
Gender	.589	0.825 (0.411-1.661)
BMI	.725	0.985 (0.906-1.071)
Diabetes condition	.005*	2.760 (1.349-5.631)
Hypertension condition	.933	0.969 (0.463-2.031)
Tumor size	.002*	1.030 (1.011-1.050)	.019*	1.028(1.005-1.053)
Lesion morphology	.445	0.761 (0.378-1.530)
TNM stage	.012*	2.507 (1.223-5.140)	.023*	2.846 (1.154-7.023)
Acute obstructive pancreatitis	.124	1.747 (0.858-3.560)
Vascular invasion	.487	1.290 (0.629-2.640)
Nerve invasion	.001*	3.450 (1.628-7.311)	.026*	3.078 (1.146-8.268)
CA19-9	.007*	1.007 (1.002-1.010)
CEA	.010*	1.008 (1.002-1.011)	.014*	1.009(1.002-1.018)
Serum amylase	.094	1.008 (0.999-1.020)
Advanced hospital surgery	.017*	0.362 (0.157-0.833)	.012*	0.286(0.109-0.756)

Abbreviations: PBTC, pancreatic body tail cancer; TNM, tumor, node, and metastasis; CA19-9, carbohydrate antigen 19-9; CEA, carcinoembryonic antigen; BMI, body mass index.

*P < .05.

OPEN IN VIEWER

Table 3.

Logistic Regression Analysis Results of CT Radiomics Model Based on CT Texture Analysis for Predicting PBTC Prognoses.

CT radiomics model	Univariate analysis		Multivariate analysis
Factors	P	Hazard ratio	P	Hazard ratio
CT enhancement mode	.027*	2.220 (1.094-4.510)
Dilation of the main pancreatic duct	.002*	1.253 (1.086-1.440)	.001*	1.420 (1.171-1.720)
Pancreatic atrophy	.217	0.643 (0.319-1.301)
Pancreatic cyst	.865	1.060 (0.526-2.150)
Peripancreatic lymph node swelling	.061	2.308 (0.963-5.530)
Peripancreatic fluid	.040*	2.154 (1.037-4.471)
Radscore1	.001*	2.093 (1.370-3.205)
Radscore2	.001*	3.281 (2.180-4.930)	.001*	4.090 (2.301-7.290)
Radscore3	.001*	2.558 (1.620-4.032)

Abbreviations: PBTC, pancreatic body tail cancer; CT, computed tomography.

*P < .05.

OPEN IN VIEWER

Table 4.

Logistic Regression Analysis Results of Combined Model Based on Above Risk Factor for Predicting PBTC Prognoses.

Combined model	Univariate analysis		Multivariate analysis
Factors	P	Hazard ratio	P	Hazard ratio
Age	.022*	1.036(1.005-1.070)
Diabetes condition	.005*	2.760 (1.349-5.631)
Tumor size	.002*	1.030 (1.011-1.050)
TNM stage	.012*	2.507 (1.223-5.140)
nerve invasion	.001*	3.450 (1.628-7.311)	.022*	7.354 (1.335-40.494)
CA19-9	.007*	1.007 (1.002-1.010)
CEA	.010*	1.008 (1.002-1.011)
Advanced hospital surgery	.017*	0.362 (0.157-0.833)	.014*	0.173 (0.042-0.703)
CT enhancement mode	.027*	2.220(1.094-4.510)
Dilation of the main pancreatic duct	.002*	1.253 (1.086-1.440)	.003*	1.446 (1.136-1.838)
Peripancreatic fluid	.040*	2.154 (1.037-4.471)
Radscore1	.001*	2.093 (1.370-3.205)
Radscore2	.001*	3.281 (2.180-4.930)	.001*	5.468 (2.177-13.732)
Radscore3	.001*	2.558 (1.620-4.032)

Abbreviations: PBTC, pancreatic body tail cancer; CT, computed tomography; TNM, tumor, node, and metastasis; CA19-9, carbohydrate antigen 19-9; CEA, carcinoembryonic antigen.

*P < .05.

Discussion

The incidence rate of pancreatic cancer in China is as high as 5.1/100 000 due to the acceleration of China's industrialization process, food pollution, high living pressure, and other factors, but the median survival period is only 4 to 6 months. With the development and wide application of imaging technology, the detection rate of pancreatic tumors keeps rising, about 5% to 10% of patients with jaundice and emaciation were finally confirmed pancreatic cancer. PBTC is a common type of pancreatic tumor with variable nature, insidious onset, and prone to be missed diagnosed or misdiagnosed.^18,19 The early diagnosis of PBTC is very challenging, and the mortality rate keeps high, which is related to its unrepresentative imaging features, rapid progression, and high metastasis rate. Also, in the clinical diagnosis and treatment practice, most PBTC patients are diagnosed in the middle or late stages; Usually, the tumors may have already infiltrated the surrounding organs, blood vessels, nerve plexus, and peripancreatic adipose tissue. Currently, surgical resection and chemotherapy combined with immunotherapy remains the only means of curing PBTC or improving survival. Although it improves the quality/rate of life of patients to a certain extent, ultimately peripheral recurrence and metastasis are common. Therefore, the prognosis prediction of PBTC is extremely important for its postoperative treatment management. It is reported that accurate pancreatic tumor information (size, location, surrounding tissue invasion, lymph node metastasis, etc) and pathological staging play a decisive role in the postoperative survival time of PBTC patients. Moreover, the new emerging radiomics technology has a good predictive effect on the prognosis of many other tumors, which is also applicable to the PBTC. Therefore, accurate preoperative CT/MR imaging details, diagnosis, and texture parameters are of great value for the prognosis of PBTC. This important information can promote the change of clinical decision-making of PBTC patients, determine the operation mode, radiotherapy and chemotherapy options, and even timely transfer to the clinical palliative care wards, which is extremely important for the high-quality late survival of patients.^20,21

Generally, the prognosis of PBTC also depends on the biological properties of the tumor itself. It has been reported that the prognosis of PBTC was related to TNM stage, degree of differentiation, histological type, surgical method, number of regional lymph node dissections, and radiotherapy and chemotherapy.^7,22,23 To reduce bias, the PBTC cases included in this study were all ductal adenocarcinomas. In addition to the above-influencing factors, we also revealed that age, diabetes, CT enhancement mode, nerve invasion, advanced hospital surgery, tumor size, CEA, CA199, dilation of the main pancreatic duct, peripancreatic fluid, radscore1/2/3 were important factors affecting the fast progression-free survival rate of PBTC. It is well known that the immunity and physical function of elderly patients are significantly lower than those of young adults, so postoperative recovery is poor with many complications; In this study, the survival time of the younger patients was significantly higher than that of the older patients, which also confirmed this view. The high blood glucose concentration and poor immune function in patients with diabetes may be related to the recurrence of PBTC, whose wound healing is slow, and complications are even more, which worsens survival. The rapid enhancement of the tumor may be related to its multibranch blood supply and microvascular invasion, which may lead to recurrence and death. In this study, 69.93% of PBTC was accompanied by peripheral nerve invasion. To delay the recurrence, general radiotherapy and chemotherapy were required after the operation, which would further damage the body function and cause obvious pain/hematochezia in patients, resulting in a poor prognosis. Surgery in top hospitals was newly discovered recurrence and poor prognosis-related factors in this study. We found that the mortality risk of da Vinci surgical robot/advanced hospital surgery combined with chemotherapy and immunotherapy for PBTC was 0.45 times that of classical surgery and symptomatic treatment and 0.22 times that of interventional therapy with chemotherapy. Of course, surgery is still the first-line treatment for a good prognosis of PBTC in this study. Besides, 67.83% of PBTC patients with good economic conditions went to the top hospitals in big cities like Beijing and Shanghai and received the da Vinci operation, heavy-ion cancer therapy, and specialized/customized ICU care and achieved significantly improved survival rate. Although the unequal treatment caused by patients’ selection bias leads to a longer survival time, which will have a certain bias on the study, this factor will not cause a large bias on the study results, because its independent AUC is only 0.601. The size of the lesion determined the difficulty of the operation and varies the possibility of unavoidable pancreatic leakage, postoperative hemorrhage, and retroperitoneal infection, while the recurrence of PBTC in patients with larger tumors is still concerning. There 16.78% of PBTC patients with significantly elevated CEA levels had liver and intestines metastases, therefore, the recurrence and death rates were high after surgical resection. The degree of expansion of the main pancreatic duct reflects the tumor's space-occupying effect or invasion and compression of the pancreatic duct and surrounding tissues of the pancreas. Peripancreatic effusion reflects the degree of tumor necrosis and degeneration or peripheral carcinomatous lymphangitis to a certain extent, thus increasing the difficulty of surgery, leading to adverse events such as pancreatic leakage and continuous bleeding, It may be doomed to a bad prognosis. In this study, Radscore demonstrated excellent predictive performance, which may be related to the underlying imaging texture parameters of the masses it extracted. In previous studies, Radiomic features (such as intensity, shape, texture, or wavelet) were speculated may provide reliable data support and be correlated with global sensitivity, heterogenicity, possible fat or carcinoma soft tissue replacement of typical imaging features to some extent in the future. More importantly, Radscore may refine and classify the overall density, signal of the tumor based on some special algorithms, which may be similar to the distribution of CT value, and is helpful to distinguish the nature of the lesion, such as the percentage of fat, fiber, calcification, blood vessel, etc^17,24,25; which has currently been applied in domestic pulmonary nodule AI recognition software and has received good clinical evaluation.^25,26Our team found that Radscore 1 may represent the enhancement mode of PTBC dense tissue, similar to the local tissue hardening and enhancement of pancreas after detailed analysis of texture parameters, while it also highlights the possibility of local tissue necrosis and fluid exudation, which is difficult to identify by naked eye; However, Radscore 2 seems to represent the possibility of microvascular thrombosis in the mass and indicate the degree of enhancement in the venous phase, which is different from normal pancreas or inflammatory lesions. Radscore1/2/3 represented the preoperative CT image texture parameters, which can tell the nature of the tumor, and thus is valuable for recurrence prediction.^17,23,27 And based on the combined model, a new nomogram was created. We also found that when R2 (radscore2) ≥ 6, the tumor's boundaries become blurred, the enhancement pattern changes, tissue density increases, ischemic necrosis increases, and so on. Moreover, as the R2 value increases and approaches 10, the possibility of low differentiation increases and the prognosis deteriorates.^28–30 As a universal marker of gastrointestinal tumors, CA19-9 is higher sensitive. Although it is a bad prognostic factor in this study, its AUC (0.610) and specificity (0.141) are not strong, so it has little clinical significance.

At last, we established a combined model, a clinicopathological model, and a CT radiomics model, and validated that the combined model provided the best predictive performance. Subsequently, similar results were obtained in the test set and DCA confirmed its superior clinical net benefit. The combined model has higher prediction efficiency than the previously reported nomogram clinical models (AUC: 0.976 vs 0.777). ²⁷ This may be related to the combination of clinical, pathological, and CT imaging parameters and the acquisition of information about the nature of the tumor. The combined model is helpful for the treatment and management of PBTC and improves the life quality of patients, receiving clinical recognition.

Conclusions

In conclusion, the total survival period of PBTC patients is poor, and it is much challenging to improve this situation; The combined model has a high predictive efficiency for the prognosis of PBTC (AUC: 0.976), higher than previously reported clinical model (AUC: 0.777); because it is mainly generated based on clinical data and CT radiomics, it does not increase the medical cost and patient burden, and may produce more detailed clinical treatment and CT image fusion information, which may provide support for clinical decision-making and clinical management of PBTC, and help to prolong the survival time of PBTC as much as possible.