Prognostic Evaluation Based on Dual Time 18F-FDG PET/CT Radiomics Features in Patients with Locally Advanced Pancreatic Cancer Treated by Stereotactic Body Radiation Therapy

DOI: https://doi.org/10.21203/rs.3.rs-1357369/v1

Abstract

Background

The prognosis of pancreatic cancer is very poor. We aim to explore the prognostic value of radiomics features from dual time 18F-FDG PET/CT images for locally advanced pancreatic cancer (LAPC) patients treated with stereotactic body radiation therapy (SBRT).

Materials and Methods

This retrospective study included 59 LAPC patients who received early and delayed 18F-FDG PET/CT scans before SBRT treatment. Patients were divided into a training cohort: n = 40, and a validation cohort: n = 19. A total of 1188 quantitative imaging features were extracted from dual time 18F-FDG PET/CT images. To avoid over-fitting, the univariate analysis and elastic net were used to obtain a sparse set of image features, that were applied to develop a radiomics score (Rad-score). Then the Rad-score was used to constructed prognosis models to predict the overall survival (OS) of patients, and stratify the patients. The univariate analysis and multivariate analysis were used to assess the correlation of OS. The Harrell consistency index (C-index) was used to evaluate the model. And the cross-validation was used to ensure the accuracy and robustness of the results.

Results

The Rad-score from dual time images contains 6 features, including intensity histogram, morphological, and texture features. In the validation cohort, the univariate analysis showed that the Rad-score was the independent prognostic factor (p = 0.0001, hazard ratio [HR]: 2.35), which was better than traditional imaging predictive indicators and clinical indicators, such as tumor volume, maximum standardized uptake value, Total glycolysis, and TNM staging. Among all the prognostic factors included in the multivariate analysis, the Rad-score was the only prognostic factor (p < 0.01, HR: 2.71) that was significantly associated with the OS of patients. In addition, according to cross-validation, the C-index of the prognostic model based on the Rad-score of dual-time PET/CT images is better than the model based on early and delay PET/CT images(0.745 vs 0.713 vs 0.593).

Conclusion

Quantitative analysis confirmed that the Rad-score from dual time 18F-FDG PET/CT images have a better prognostic value than the Rad-score from single-time images. It has potential significance for identifying the individualized risk of LAPC patients treated with SBRT.

Background

Pancreatic cancer is a malignant tumor. Although the prognosis of patients with pancreatic cancer patients has improved due to early diagnosis and treatment methods, the 1-year and 5-year survival rate of patients is only 15%, and 9%, respectively[1,2]. Surgical resection can effectively prolong the survival of patients with pancreatic cancer. However surgical resection can only be performed in about 20% of patients, and about 20–40% of locally advanced pancreatic cancer (LAPC) patients cannot be surgically treated, and their median survival time was only 6–14 months[3]. For these LAPC patients who cannot tolerate surgical treatment, stereotactic body radiation therapy(SBRT) was a commonly used treatment[4–5]. However, not all patients with LAPC can benefit from SBRT[6], so it is very important to predict the outcome of treatment based on the patient's information.

The 18F-FDG PET/CT image has been widely used in the staging, diagnosis, and localization of pancreatic cancer[7]. Previous studies have shown that some indicators from PET imaging have a certain prognostic value for pancreatic cancer. For example, the maximum and average standard intake values (SUVmax and SUVmean) are semi-quantitative values frequently used as reference indicators in clinical diagnosis and have been proven to predict the overall survival (OS) of patients. However, the generalizable capability of SUVmax and SUVmean was poor [8–10]. The metabolic tumor volume (MTV) and total disease glycolysis (TLG) have also been shown to be related to the OS of the patients[8,11]. But they are related to the size of the tumor and the information contained seems to be limited. For dual-time PET/CT images, Santhosh S et al[12] showed that SUVmax from early and delayed images cannot predict the OS of patients. And for the retention index (RI), the results of the study are also inconsistent with previous research results[13]. The ability of these conventional PET features to accurately distinguish different risk groups seems to be limited. Therefore, there was a need for new imaging features to predict the prognosis of pancreatic cancer patients.

Tumor heterogeneity is an important factor affecting the prognosis of patients after treatment[14]. Radiomics, which involves the extraction of quantitative features from medical images to provide a comprehensive characterization of the entire tumor, reflect the spatial relationship and heterogeneity of voxel intensities within tumors[15–17]. Currently, studies have shown that radiomics features can be used for cancer diagnosis, prognosis, and preoperative staging[18–20]. Some studies have shown that texture features from PET/CT images can be used for prognostic evaluation of pancreatic cancer patients. For example, Yue Y[21] et al. studied the relationship between texture features from PET images before and after chemotherapy and the OS of patients with pancreatic cancer. The results showed that some texture features (homogeneity, variance, sum mean, the difference before and after cluster tendency chemotherapy) were prognostic factors of patients. And, Cui[22] et al showed that texture features from early PET images showed better prognostic value than conventional PET features in patients treated with SBRT. Previous studies have shown that continuously dynamic 18F-FDG PET/CT parameters can be used as predictors of patient survival[12,23]. However, it was difficult to achieve 60-minute, dynamic 18F-FDG PET/CT scans for patient prognosis. There, we assumed that the Rad-score from dual-time static 18F-FDG PET/CT images can replace dynamic imaging to a certain extent, and predict the prognosis of patients according to the changes of tumor texture.

In this study, we hypothesized that the radiomics score (Rad-score) calculated by a linear combination of the radiomics features from dual time images could be the reference indicator for the prognosis of LAPC patients. We evaluated whether the Rad-score from dual-time imaging had a better prognostic value than the Rad-score from early or delayed images. The ultimate aim of this retrospective study was to explore the role of the radiomics features from dual time 18F-FDG PET/CT images in predicting the prognosis of LAPC patients treated with SBRT.

Results

Baseline Clinical Information of the Patients

These 59 patients were divided into a training cohort and a validation cohort at a ratio of 2:1, with 40 patients in the training cohort and 19 patients in the validation cohort. The baseline clinical information was shown in Table 1. Female patients accounted for about 36%, with an average age of 65.59 ± 10.5 years. All patients received SBRT and underwent 18F-FDG PET/CT scan before treatment. There was a significant difference in the N-stage between the training and validation cohorts (p = 0.042). There were no significant differences in other clinical characteristics.

Features Selection

To select more robust features, 216 features with ICC less than or equal to 0.75 were eliminated through inter-observer analysis. Then we incorporate the remaining 972 features into the subsequent resampling experiment. Finally, we select the 6 most frequent features to retrain the final Cox regression. The categories of these 6 features and their weights in the re-trained Cox model are given in Table 2.

Early PET/CT Images Analysis

For early image analysis, according to the Rad-score in the training cohort, patients were divided into high RS groups and low RS groups. There was a significant difference in OS between the high-risk group and the low-risk group (p < 0.001) in the training and the validation cohorts. And the Kaplan–Meier curve for the Rad-score is shown in Figure 2A,2B. In terms of the prognostic accuracy of the prognostic model, the Rad-score calculated from the early image reached 0.718 ± 0.055 in the validation cohort. 

Delayed PET/CT Images Analysis

For delayed image analysis, The Kaplan–Meier curves of the high-risk group and the low-risk group were shown in Figures 2C and 2D. The OS of patients in the low-risk group was significantly better than that of the high-risk group (p < 0.001 in training cohorts). We constructed a Cox proportional hazard regression model based on the  Rad-score in the training cohort, and its survival prediction accuracy in the validation cohort reached 0.593 ± 0.058.

Dual-time-point PET/CT images Analysis

For dual time image analysis, the Kaplan–Meier curve of the Rad-score for high-risk and low-risk groups was seen in both training and validation cohorts (Figure 2E and 2F, p < 0.001). In the training cohort, there was a significant difference in survival of time of patients between the low-risk group and those in the high-risk group, which was verified in the validation cohort (p < 0. 01 OS analysis) (Fig. 3). The C-index of the dual-time model based on the Rad-score was 0.745 ± 0.050 in the validation cohort. In addition, the results of the Wilcoxon rank-sum test show that there are significant differences between models based on dual-time images and single-time image models (p < 0.01). Furthermore, we compared the prediction accuracy of several major radiomic features used in the Rad-score and the Rad-score. The results show that the Rad-score has better prediction performance.

To determine the independent risk factors for patients treated with SBRT, we performed univariate and multivariate Cox analysis of the Rad-score from dual time images, clinical information, and conventional PET features (SUV, MTV, TLG). The results were shown in Table 3. According to univariate Cox regression analysis, the Rad-score from dual time images was significantly correlated with the OS (p < 0.001) in the validation cohort. And higher SUVmax, MTV, and TLG from the delayed images were significantly correlated with shorter OS(p = 0.026, 0.027), and clinical factors such as T-stage, Dose, and Chemotherapy were also significantly correlated with OS (all p < 0.05). On the other hand, with a multivariate Cox regression analysis, the   Rad-score from dual time images was the only independent prognostic factor (p < 0.001). 

Discussion

The purpose of this study was to explore the predictive value of radiomics features from dual-time images on the prognosis of LAPC patients treated with SBRT. We found that the Rad-score from dual time 18F-FDG PET/CT images can be used to predict the prognosis of LAPC patients after SBRT, and can achieve the prognostic stratification with OS of patients. This result has clinical significance for promoting the precise treatment of LAPC.

The radiomics has attracted a lot of attention in its ability to non-invasively analyze tumor heterogeneity; and provides a viable tool for patient prognosis. In this study, we conducted a two-stage experimental setup to explore the best model for a prognosis for patients with pancreatic cancer. First, we use 594 radiomics features extracted from early imaging or delayed imaging to develop a single-time prognostic model, and then we combine the radiomics features extracted from dual time PET/CT images to develop a dual-time prognostic model. In these two stages, the six radiomics features with the highest frequency were used to calculate the Rad-score. Univariate analysis showed that the Rad-score of both early and delayed images was significantly correlated with the OS of LAPC patients (p < 0.01, HR: 2.60 and 2.70). This result indicated that both early and delayed PET/CT images contain patient prognostic information[12,22]. The Rad-score from dual time images is also significantly correlated with OS of patients (p < 0.01, HR: 2.35). The C-index of the model based on dual time images was significantly higher than that of the model based on early and delayed images (p < 0.05, Wilcoxon rank-sum test). At the same time, it must be pointed out that there are differences in the division of high-risk and low-risk groups according to the Rad-score of single time images, while an accurate grouping can be obtained by analyzing the Rad-score of dual time images. Two representative patients were shown in Fig. 3. These results showed that dual-time PET/CT images can provide more prognostic information and offset the limitations of single-time PET/CT images[12]. This may be related to the kinetics of tumor uptake. As the uptake time increases, the uptake of 18F-FDG by malignant tumor tissues will increase significantly[31–32].

In this study, the Rad-score from dual time images includes three types of features: shape feature, first-order gray statistics feature, and texture feature. Different types of texture features can reflect the inherent heterogeneity of tumors from different angles[33–34]. In addition, the optimal features include both original texture features and texture features based on wavelet transform (wavelet coefficients: LLH + LHL + HLL, HHH (HHH = highpass filter + highpass filter + highpass filter)), which reflect the heterogeneity of tumors on different spatial scales. On the one hand, it must be pointed out that when these six radiomics features were used alone, none of the features can predict patient survival better than the Rad-score, indicating the complementarity of information between different features. On the other hand, these characteristics are also significantly related to the OS of patients. The solidity reflects the complexity of the homogenous region of the tumor. The smaller the value, the higher the complexity of the tumor and the worse the prognosis of the patients[35]. The gray level difference statistics (GLDS) calculates the contrast of the image and reflects the roughness of the texture. The Contrast [36] (PET, GLDS, HHH) shows that the metabolic changes in the lesions in PET images have a strong resolution, the functional metabolic changes in the lesions are larger and the texture is coarser in patients with poor prognosis. The Busyness (Neighborhood Gray-Tone Difference Matrix (NGTDM), PET) measures the change from pixel to the adjacent pixel. A high value of business indicates that the intensity between a pixel and its neighborhood changes rapidly, indicating that the more complex the tumor is in patients with poor prognosis. The Correlation and The Energy from Gray Level Size Zone Matrix (GLSZM) in CT images quantify the degree of non-uniformity of the gray level in images. The patients with a better prognosis showed better texture consistency.

For LAPC patients, the median survival was 4–16 months. The clinicopathological parameters, including T-stage, chemotherapy, and dose, were found to be strong predictors of prognosis in LAPC patients receiving SBRT in univariate Cox analysis. However, the above indicators did not show prognostic value for patients in the multivariate Cox analysis study. For the conventional PET features, the TLG from delayed PET/CT images is significantly associated with a poor prognosis. This reaffirms the fact that the metabolic tumor volume combined with the tumor range is a better predictor of patient survival. However, in this study, the TLG of the early images does not correlate with the OS. One possible reason is that compared with the early images, delayed images may better reflect the uptake of 18F-FDG by malignant tumors. This result is different from our previous research[8]. One possible reason is that the tumor contour is outlined in different ways, and another possible reason is the difference in the number of patients.

There are some limitations to this study. First, the samples in this study were from a single-center, the sample size available for analysis was small, and the potential of selection bias cannot be ruled out, which limits the accuracy and reliability of the results. Therefore, we hope that the results of this study can be repeated using larger data sets and multiple centers in the future. Second, the ROI/VOI was drawn manually, which is very time-consuming and inconvenient, and the predicted performance may be sensitive to the ROI/VOI depicting pancreatic lesions. In future research, automatic segmentation or semi-automatic segmentation could be achieved through the application of deep learning. Finally, PET/CT images acquisition was carried out in the presence of respiratory movement, which may distort the real metabolic activity and affect the accurate segmentation of tumors. In future research, the respiratory gating technique may help to improve the quantitative accuracy of PET imaging.

Conclusion

The Rad-score obtained from dual time 18F-FDG PET/CT images reflect the heterogeneity of intertumoral metabolism from different aspects. It is a powerful predictor of survival for patients with locally advanced pancreatic cancer treated with SBRT. The radiomics analysis of dual-time PET/CT images can help patients choose the appropriate treatment plan and realize precision medicine.

Materials And Methods

Patients

This retrospective study was approved by the Ethics Committee of Changhai Hospital, and informed consent was given to all participants. The criteria for patient inclusion were: (a) confirmation of pancreatic cancer on pathological examination of the patient after PET/CT scan; (b) available dual time 18F-FDG PET/CT images; and (c) underwent SBRT treatment. The exclusion criteria were: (a) other malignant tumors; (b) death due to diseases other than pancreatic cancer during the follow-up period; and (c) a metal positioning mark implanted in the tumor lesion. Finally, a total of 59 patients who underwent dual-time 18F-FDG PET/CT examinations in our hospital between January 2012 and January 2018 were identified and included in the study. Some of the data in this study have been reported[8]. 

PET/CT imaging protocols

The dual time 18F-FDG PET/CT images data of the patients was collected on a Biograph tripoint 64-layer 52-ring HD PET/CT scanner (Siemens, Germany). Before the whole-body scan, the patients were required to fast for at least 6 hours. When their blood sugar was lower than 11.1mmol/L, 18F-FDG at the dose of 3.70 ~ 5.55MBq/kg was injected, and the early scanning was started 50-60 minutes after the injection. The whole-body PET scan covers 5-6 beds, with an acquisition time per bed of about 2.5 minutes, a spatial resolution of 4.07 × 4.07 mm2, and a scan thickness of 3mm. The parameters of the CT scan were a current of 170 mA, a voltage of 120kv, a spatial resolution of 0.98 × 0.98 mm2, and a scan thickness of 3 mm over a time of 18.67-21.93 s. The PET and CT image matrix size were 168 × 168 and 512 × 512, respectively. After 120-150 minutes, delayed scanning was started. The delayed PET/CT images only contain the head to tail of the pancreas. Patients were required to breathe shallowly during PET/CT scans to reduce the impact of breathing exercises.

Image analysis

The radiomics workflow is shown in Figure 1. For image preprocessing, we used the 3D Slicer (version 4.10.2) to resample the original PET image and co-register it with the corresponding CT images[24-25]. Under the guidance of the PET images, two radiologists with more than 10 years of clinical experience outlined the tumor contour on the CT images. Then the voxels of the CT were clipped to [−10, 100] Hounsfield Units to reduce the interference of fat and other factors on texture features[26-27]. For the PET images, a classical normalization factor (body weight) was used to convert the voxel values to SUV values[28], and a square root transformation was used to reduce noise. Finally, the voxel values of PET and CT images were normalized to [0, 255]. 

The feature extraction algorithm was applied using MATLAB (The MathWorks, Inc. Natick, MA, USA). Before extracting the three-dimensional radiomics features, firstly, the volumes of interests (VOIs) were obtained by performing nearest cubic trilinear interpolation on all ROI areas in the Z-axis direction. Then the spatial resolution of CT and PET images was resampled to 1 × 1 × 1 mmby cubic trilinear interpolation. Finally, a total of 594 features were extracted from the early and the delayed PET/CT images respectively, including 3 groups: texture features, shape features, and wavelet features (Supplementary Table I).

Feature Selection

Before features analysis, each radiomics feature value was independently normalized by subtracting its mean and then dividing by its standard deviation. To improve the reproducibility and robustness of the optimal features, the intraclass correlation coefficient (ICC) and the bootstrap (100 times, 40 cases of data/each time) were used for feature selection. In resampling, the specific process of filtering each feature set is as follows: First, the features with p-value <0.05 in the univariate analysis were included in the elastic network[29]. Then cross-validation was used to find the optimal parameters of the elastic network. We repeated this process 100 times and recorded the selected features each time. 

To calculated the Rad-score, we counted the frequency of the features and included the top six features into the final feature set. Then we retrain the multivariate Cox regression model[30] with the final feature subset in the training cohort and get the corresponding weights. The Rad-score is defined as the sum of the weights of each feature, which will be used in subsequent related experiments.

Statistics analysis

Statistical analysis was implemented in R (version 3.6.3) software. First, the patients were divided into a high-risk group and a low-risk group using the median of the Rad-score as a cut-off value. And the Kaplan-Meier survival analysis with log-rank test was used to compare the low-risk and the high-risk groups. Then the clinical features, conventional PET imaging features, and the Rad-score were included in the univariate and multivariate Cox regression analysis. In addition, the Rad-score was also used to build the prognosis model based on Cox proportional hazard regression analysis and C-index was used to evaluate the model. And the three-fold cross-validation was used to evaluate the performance of survival prediction. We evaluate the difference in C-index between the two indicators based on the Wilcoxon rank-sum test. 

Declarations

Conflict of interest: The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.

Compliance with Ethical Standards. This retrospective study was approved by the Institutional Review Board of the Changhai hospital. Inform consent was waived.

Authors Contributions. ZZY, CJZ, and CC conceptualized and designed the study. FW and SNR carried out relevant experiments and drafted the manuscript. SNR, CC, and TW carried out data annotation. XDY, ZYW, and ZBL revised the manuscript. All authors contributed to the article and approved the submitted version.

Funding. This study is supported by the National Natural Science Foundation of China #81871390 and #62101551, the "Climbing" program-234 plan of Changhai hospital #2019YPT002, and the Starting Foundation for Young Researcher of Changhai hospital #2019QNB07.

Availability of data and materials. The data sets used and/or analyzed during the current study are available from the corresponding author on reasonable request.

References

  1. Siegel RL, Miller KD, Jemal AJCACJfC. Cancer statistics, 2018. 2018;60(suppl 12):277–300.
  2. Bray F, Ferlay J, Soerjomataram I, Siegel RL, Torre LA, Jemal AJCacjfc. Global cancer statistics 2018: GLOBOCAN estimates of incidence and mortality worldwide for 36 cancers in 185 countries. CA-Cancer J. Clin. 2018;68(6):394–424.
  3. Tempero MA, Malafa MP, Al-Hawary M, Asbun H, Bain A, Behrman SW, Benson AB, Binder E, Cardin DB, Cha C, Chiorean EG. Pancreatic adenocarcinoma, version 2.2017: Clinical practice guidelines in Oncology. J. Natl. Compr. Cancer Netw. 2017;15(8):1028.
  4. Petrelli F, Comito T, Ghidini A, Torri V, Scorsetti M, Barni S. Stereotactic body radiation therapy for locally advanced pancreatic cancer: a systematic review and pooled analysis of 19 trials. Int. J. Radiat. Oncol. Biol. Phys. 2017 Feb 1;97(2):313 − 22.
  5. Balaban EP, Mangu PB, Khorana AA, Shah MA, Mukherjee S, Crane CH, Javle MM, Eads JR, Allen P, Ko AH, Engebretson A. Locally advanced, unresectable pancreatic cancer: American Society of Clinical Oncology clinical practice guideline. J. Clin. Oncol. 2016 Aug 1;34(22):2654-68.
  6. De Bari B, Porta L, Mazzola R, Alongi F, Wagner AD, Schäfer M, Bourhis J, Ozsahin M. Hypofractionated radiotherapy in pancreatic cancer: Lessons from the past in the era of stereotactic body radiation therapy. Crit. Rev. Oncol./Hematol. 2016 Jul 1;103:49–61.
  7. Bang S, Chung HW, Park SW, Chung JB, Yun M, Lee JD, Song SY. The clinical usefulness of 18-fluorodeoxyglucose positron emission tomography in the differential diagnosis, staging, and response evaluation after concurrent chemoradiotherapy for pancreatic cancer. J. Clin. Gastroenterol. 2006 Nov 1;40(10):923-9.
  8. Ren S, Zhu X, Zhang A, Li D, Zuo C, Zhang H. Prognostic value of 18F-FDG PET/CT metabolic parameters in patients with locally advanced pancreatic cancer treated with stereotactic body radiation therapy. Cancer Imaging. 2020 Dec;20(1):1–1.
  9. Su W, Ren S, Zhu X, Zhang H, Zuo C. Standardized thresholds of volume-based PET/CT parameters predicting survival of patients with pancreatic head cancer treated with stereotactic body radiation therapy. Ann. Nucl. Med. 2020;34(6):379 − 87.
  10. Zhang A, Ren S, Yuan Y, Li X, Zhu X, Jiang L, et al. Prognostic values of F-18-FDG PET/CT metabolic parameters and clinical figures in locally advanced pancreatic cancer underwent chemotherapy combined with stereotactic body radiation therapy. Medicine (Baltimore). 2019;98(13).
  11. Dholakia AS, Chaudhry M, Leal JP, Chang DT, Raman SP, Hacker-Prietz A, Su Z, Pai J, Oteiza KE, Griffith ME, Wahl RL. Baseline metabolic tumor volume and total lesion glycolysis are associated with survival outcomes in patients with locally advanced pancreatic cancer receiving stereotactic body radiation therapy. Int. J. Radiat. Oncol. Biol. Phys. 2014 Jul 1;89(3):539 − 46.
  12. Santhosh S, Mittal BR, Bhasin D, Rana SS, Bhattacharya A, Srinivasan R, Nada R, Gupta R. Dual-phase 18F-FDG PET/CT imaging in the characterization of pancreatic lesions: does it offer prognostic information? Nucl. Med. Commun. 2014 Oct 1;35(10):1018-25.
  13. Lyshchik A, Higashi T, Nakamoto Y, Fujimoto K, Doi R, Imamura M, Saga T. Dual-phase 18 F-fluoro-2-deoxy-D-glucose positron emission tomography as a prognostic parameter in patients with pancreatic cancer. Eur. J. Nucl. Med. Mol. Imaging. 2005 Apr;32(4):389 − 97.
  14. Bedard PL, Hansen AR, Ratain MJ, Siu LL. Tumour heterogeneity in the clinic. Nature. 2013 Sep;501(7467):355 − 64.
  15. Chicklore S, Goh V, Siddique M, Roy A, Marsden PK, Cook GJ. Quantifying tumor heterogeneity in 18 F-FDG PET/CT imaging by texture analysis. Eur. J. Nucl. Med. Mol. Imaging. 2013 Jan;40(1):133 − 40.
  16. Lambin P, Leijenaar RT, Deist TM, Peerlings J, De Jong EE, Van Timmeren J, Sanduleanu S, Larue RT, Even AJ, Jochems A, van Wijk Y. Radiomics: the bridge between medical imaging and personalized medicine. Nat. Rev. Clin. Oncol. 2017 Dec;14(12):749 − 62.
  17. Gillies RJ, Kinahan PE, Hricak H. Radiomics: images are more than pictures, they are data. Radiology. 2016 Feb;278(2):563 − 77.
  18. Liu Z, Li M, Zuo C, Yang Z, Yang X, Ren S, Peng Y, Sun G, Shen J, Cheng C, Yang X. Radiomics model of dual-time 2-[18 F] FDG PET/CT imaging to distinguish between pancreatic ductal adenocarcinoma and autoimmune pancreatitis. Eur. Radiol. 2021 Mar 6:1–9.
  19. Klaassen R, Larue RT, Mearadji B, van der Woude SO, Stoker J, Lambin P, van Laarhoven HW. Feasibility of CT radiomics to predict treatment response of individual liver metastases in esophagogastric cancer patients. PLoS One. 2018 Nov 15;13(11):e0207362.
  20. Luo HS, Huang SF, Xu HY, Li XY, Wu SX, Wu DH. A nomogram based on pretreatment CT radiomics features for predicting complete response to chemoradiotherapy in patients with esophageal squamous cell cancer. Radiat. Oncol. 2020 Dec;15(1):1–1.
  21. Yue Y, Osipov A, Fraass B, Sandler H, Zhang X, Nissen N, Hendifar A, Tuli R. Identifying prognostic intratumor heterogeneity using pre-and post-radiotherapy 18F-FDG PET images for pancreatic cancer patients. J. Gastrointest. Oncol. 2017 Feb;8(1):127.
  22. Cui Y, Song J, Pollom E, Alagappan M, Shirato H, Chang DT, Koong AC, Li R. Quantitative analysis of 18F-fluorodeoxyglucose positron emission tomography identifies novel prognostic imaging biomarkers in locally advanced pancreatic cancer patients treated with stereotactic body radiation therapy. Int. J. Radiat. Oncol. Biol. Phys. 2016 Sep 1;96(1):102-9.
  23. Epelbaum R, Frenkel A, Haddad R, Sikorski N, Strauss LG, Israel O, Dimitrakopoulou-Strauss A. Tumor aggressiveness and patient outcome in cancer of the pancreas assessed by dynamic 18F-FDG PET/CT. J. Nucl. Med. 2013 Jan 1;54(1):12 − 8.
  24. Mattes D, Haynor DR, Vesselle H, Lewellen TK, Eubank W. PET-CT image registration in the chest using free-form deformations. IEEE Trans. Med. Imaging. 2003 Apr 2;22(1):120-8.
  25. Pieper S, Halle M, Kikinis R. 3D Slicer. In2004 2nd IEEE international symposium on biomedical imaging: nano to macro (IEEE Cat No. 04EX821) 2004 Apr 18 (pp. 632–635). IEEE.
  26. Westphalen AC. Diagnosis of renal angiomyolipoma with CT hounsfield unit thresholds. Radiology. 2012 Jan;262(1):370-1.
  27. Mathur A, Hernandez J, Shaheen F, Shroff M, Dahal S, Morton C, Farrior T, Kedar R, Rosemurgy A. Preoperative computed tomography measurements of pancreatic steatosis and visceral fat: prognostic markers for dissemination and lethality of pancreatic adenocarcinoma. Hpb. 2011 Jun 1;13(6):404 − 10.
  28. Kubota K, Matsuzawa T, Ito M, Ito K, Fujiwara T, Abe Y, Yoshioka S, Fukuda H, Hatazawa J, Iwata R, Watanuki S. Lung tumor imaging by positron emission tomography using C-11 L-methionine. J. Nucl. Med. 1985 Jan 1;26(1):37–42.
  29. Tibshirani R. The lasso method for variable selection in the Cox model. Stat. Med. 1997 Feb 28;16(4):385 − 95.
  30. Aerts HJ, Velazquez ER, Leijenaar RT, Parmar C, Grossmann P, Carvalho S, Bussink J, Monshouwer R, Haibe-Kains B, Rietveld D, Hoebers F. Decoding tumour phenotype by noninvasive imaging using a quantitative radiomics approach. Nat. Commun. 2014 Jun 3;5(1):1–9.
  31. Chen S, Harmon S, Perk T, Li X, Chen M, Li Y, Jeraj R. Diagnostic classification of solitary pulmonary nodules using dual time 18 F-FDG PET/CT image texture features in granuloma-endemic regions. Sci Rep. 2017 Aug 24;7(1):1–8.
  32. Parghane RV, Basu S. Dual–time point 18F-FDG-PET and PET/CT for differentiating benign from malignant musculoskeletal lesions: opportunities and limitations. InSeminars in nuclear medicine 2017 Jul 1 (Vol. 47, No. 4, pp. 373–391). WB Saunders.
  33. Jiang Y, Chen C, Xie J, Wang W, Zha X, Lv W, Chen H, Hu Y, Li T, Yu J, Zhou Z. Radiomics signature of computed tomography imaging for prediction of survival and chemotherapeutic benefits in gastric cancer. EBioMedicine. 2018 Oct 1;36:171 − 82.
  34. Hatt M, Majdoub M, Vallières M, Tixier F, Le Rest CC, Groheux D, Hindié E, Martineau A, Pradier O, Hustinx R, Perdrisot R. 18F-FDG PET uptake characterization through texture analysis: investigating the complementary nature of heterogeneity and functional tumor volume in a multi–cancer site patient cohort. J. Nucl. Med. 2015 Jan 1;56(1):38–44.
  35. Yoon HJ, Park H, Lee HY, Sohn I, Ahn J, Lee SH. Prediction of tumor doubling time of lung adenocarcinoma using radiomic margin characteristics. Thorac. Cancer. 2020 Sep;11(9):2600-9.
  36. Tegos TJ, Mavrophoros D, Sabetai MM, Elatrozy TS, Dhanjil S, Karapataki M, Witt N, Nicolaides AN. Types of neurovascular symptoms and carotid plaque ultrasonic textural characteristics. J. Ultrasound Med. 2001 Feb;20(2):113 − 21.

Tables

Table 1. Clinical characteristics of the patients in the training cohort and the validation cohort. 

Variable

Training cohort(n=40)

Validation cohort(n=19)

P value

Age(years)*

68(42, 82)

65(50, 76)

0.355

Sex#

 

 

0.890

Male

26(65%)

12(63.2%)

 

Female

14(35%)

7(36.8%)

 

T stage#

 

 

0.826

1

1(2.5%)

1(5.3%)

 

2

11(27.5%)

4(21.1%)

 

3

15(37.5%)

7(36.8%)

 

4

13(32.5%)

7(36.8%)

 

N stage#

 

 

0.042

0

26(65%)

7(36.8%)

 

1

14(35%)

12(63.2%)

 

ECOG#

 

 

0.826

0

8(20%)

5(26%)

 

1

18(45%)

6(32%)

 

2

14(35%)

8(42%)

 

CA19-9*

304.45(2, 1200)

186.1(2, 2125)

0.987

Longest Diameter(cm)*

3.65(1.7, 7.5)

3.7(1, 6.5)

0.727

Location#

 

 

0.944

Head

27(67.5%)

12(63%)

 

Body/Distal

13(32.5%)

7(37%)

 

Chemotherapy#

 

 

0.944

0

27(67.5%)

13(68.4%)

 

1

13(32.5%)

6(31.6%)

 

Dose*

37(30.4,46.8)

37.2(30.0,46.8)

0.733

OS(month)*

14.3(7.5, 56.8)

15.6(7.5,43.9)

0.655

#Data are the number of patients, and data in parentheses are the ratio. *Data are the median, and data in parentheses are the range. Chi-square test and Mann-Whiney U test are used to compare the difference between categorical and continuous variables in the training cohort and the validation cohort, respectively. ECOG Eastern Cooperative Oncology Group; CA19-9 carbohydrate antigen 19-9; Dose Radiotherapy Dose (Gy). 

Table 2. Radiomics features were selected via the elastic net and the corresponding coefficients in retraining the Cox regression model. 

Feature Name

Categories

Modality

Time

Weights

Early

Delay

Dual

Solidity

Shape

-

Early

-3.55

-

-3.15

Contrast

GLDS(HHH)

CT

3.00

3.51

Contrast

GLDS(HHH)

PET

1.86

3.84

Energy

GLDS(LLL)

CT

3.44

 

Contrast

GLDS(HHH)

PET

0.03

 

Energy

GLCM(LLH-LHL-HLL)

CT

-1.01

 

 

 

 

 

 

 

 

Busyness

NGTDM(LLL)

PET

Delay

-

0.86

0.62

Entropy

Histogram(Original)

CT

-1.74

-4.53

Gray level non-uniformity

GLSZM(LLH-LHL-HLL)

CT

2.67

-0.01

Busyness

NGTDM(Original)

PET

-0.77

 

Entropy

GLDS(Original)

CT

-0.37

 

Mean

GLDS(LLL)

PET

-0.59

 

Rad_score = image


Table 3. Univariate and multivariate regression analysis for Rad-score, clinical risk factors, and conventional PET features in the validation cohort.

Parameters

Univariate analysis

Multivariate analysis

HR (95%CI)

p-value

HR(95%CI)

p-value

Age

1 (0.98-1)

0.5

 

 

Sex

1.2 (0.66-2.1)

0.61

 

 

ECOG

0.86 (0.59-1.3)

0.44

 

 

Tumor Diameter

1.2 (0.97-1.5)

0.1

 

 

Location

1.1 (0.62-2)

0.72

 

 

T stage

1.7 (1.2-2.3)

0.002

1.402 (0.94-2.09)

0.09

N stage

1 (1-1)

0.14

 

 

CA19-9

1.5 (0.88-2.6)

0.057

 

 

Chemotherapy

0.45 (0.24-0.84)

0.012

0.61 (0.28-1.31)

0.21

Dose

0.92 (0.86-0.97)

0.004

0.95 (0.891.07)

1.01

SUVmax (early)

2.3 (0.77-7.1)

0.13

 

 

SUVmean (early)

2.3 (0.78-6.8)

0.13

 

 

MTV(early)

1.3 (0.32-5)

0.74

 

 

TLG (early)

2.5 (0.64-10)

0.18

 

 

SUVmax (delay)

4.3 (1.2-16)

0.025

8.17(0.7-95)

0.09

SUVmean(delay)

2.9 (0.88-9.8)

0.08

 

 

MTV (delay)

6.7 (1.4-32)

0.019

-

0.41

TLG (delay)

7.4 (1.6-33)

0.009

-

0.42

Rad_score(dual)

2.6 (2-3.5)

<0.001

2.8 (2.05-3.80)

<0.001