Radiomics Nomogram for Prediction of Peritoneal Metastasis in Patients With Gastric Cancer

Weicai Huang, Kangneng Zhou, Yuming Jiang, Chuanli Chen, Qingyu Yuan, Zhen Han, Jingjing Xie, Shitong Yu, Zepang Sun, Yanfeng Hu, Jiang Yu, Hao Liu, Ruoxiu Xiao, Yikai Xu, Zhiwei Zhou, Guoxin LiFrontiers in Oncology 2020 · DOI

TL;DR

This study introduces a novel radiomics-based nomogram that uses CT scan data to predict peritoneal metastasis (PM) in gastric cancer (GC) patients before surgery. The nomogram, which combines radiomics features with clinical T and N stages, demonstrates promising accuracy in identifying patients at high risk for PM, potentially improving preoperative decision-making and treatment planning.

What Problem Does This Paper Address?

Gastric cancer (GC) is a significant global health challenge, and peritoneal metastasis (PM) is a critical factor associated with poor prognosis, particularly in advanced stages. Accurately assessing PM status before surgery is crucial for guiding treatment decisions, planning adjuvant therapies, and avoiding unnecessary surgical interventions. While CT scans are commonly used to detect PM, their sensitivity can be limited, leading to missed diagnoses. Laparoscopy is the gold standard for diagnosis but is invasive and not suitable for all patients. Therefore, there's a pressing need for a non-invasive, accurate method to predict PM status preoperatively.

Methods

The researchers employed a radiomics approach, which involves extracting a large number of quantitative features from medical images (in this case, CT scans) to uncover hidden patterns.

  • Feature Extraction: They extracted 292 quantitative image features from CT scans of GC patients.
  • Feature Selection: To identify the most relevant features for predicting PM, they used the Least Absolute Shrinkage and Selection Operator (LASSO) logistic regression model. This method helps to reduce the dimensionality of the data and select a parsimonious set of predictors while preventing overfitting.
  • Radiomics Signature Development: After feature selection and addressing collinearity, four key radiomics features were identified and combined to create a "radiomics signature" or "Rad-score."
  • Prediction Model Development: A multivariable logistic regression model was used to build a prediction model incorporating the radiomics signature.
  • Nomogram Construction: To create a more clinically applicable tool, they developed a radiomics nomogram by integrating the radiomics signature with established clinical factors: the clinical T stage (cT) and N stage (cN) of the tumor.
  • Validation: The developed model and nomogram were rigorously validated using two independent cohorts (internal and external) to assess their generalizability and robustness. Performance was evaluated using calibration (how well predicted probabilities match actual outcomes) and discrimination (how well the model distinguishes between PM-positive and PM-negative cases) via Receiver Operating Characteristic (ROC) curves and Area Under the Curve (AUC) metrics.

Datasets

The study utilized retrospective data from three independent cohorts of gastric cancer patients:

  • Training Cohort: 562 patients from Sun Yat-sen University Cancer Center.
  • Internal Validation Cohort: 106 patients from Sun Yat-sen University Cancer Center.
  • External Validation Cohort: 287 patients from Nanfang Hospital of Southern Medical University.

In total, 955 patients were included. All patients had pathological T4 stage GC and underwent contrast-enhanced abdominal CT scans within four weeks of surgery. The study focused on patients with T4 stage GC because they are at a higher risk of peritoneal metastasis.

Key Insights and Ideas

  • Radiomics as a "Digital Biopsy": The study highlights radiomics' potential to extract rich, quantitative information from standard CT scans, acting as a non-invasive "digital biopsy" that goes beyond visual interpretation.
  • Predictive Power of Texture Features: The fact that a combination of just four texture-based radiomics features could significantly predict PM status is a key insight. It suggests that subtle, complex patterns within the tumor's imaging characteristics hold crucial prognostic information.
  • Synergy of Radiomics and Clinical Factors: The nomogram's superior performance when combining radiomics with clinical T and N stages underscores the importance of integrating advanced imaging biomarkers with established clinical parameters for a more comprehensive predictive model.
  • Clinical Utility Beyond Diagnosis: The nomogram isn't just for diagnosis; it aims to provide a probability of PM, which can directly inform treatment decisions, such as the need for neoadjuvant chemotherapy or more aggressive surgical approaches.

Key Results

  • Radiomics Signature Performance: The radiomics signature (Rad-score) alone showed good predictive performance, with AUCs ranging from 0.745 to 0.802 across the training and validation cohorts.
  • Nomogram Performance: The radiomics nomogram, incorporating the radiomics signature, cT stage, and cN stage, demonstrated even better predictive accuracy. AUCs for the nomogram were:
  • Training cohort: 0.792 (95% CI, 0.748-0.836)
    • Internal validation cohort: 0.870 (95% CI, 0.795-0.946)
    • External validation cohort: 0.815 (95% CI, 0.763-0.867)
  • Independent Predictor: The radiomics signature was identified as an independent predictor of peritoneal metastasis in multivariable logistic analysis.
  • Clinical Usefulness: Decision curve analysis (DCA) indicated that the nomogram provided a net benefit for predicting PM status across various threshold probabilities, suggesting its clinical utility.

Limitations and Future Work

  • Retrospective Nature and Single-Institution Bias: The study relied on retrospective data from two Chinese institutions. The authors acknowledge that these findings need validation in larger, multicenter, prospective studies to establish higher-level evidence.
  • Clinical vs. Pathological Stage Discrepancy: The model is primarily designed for patients with preoperative T3 or T4 stage diagnoses. However, inconsistencies between clinical and pathological staging can occur, potentially leading to missed opportunities for using this model in some patients.
  • Exclusion of Other Biomarkers: Certain serological markers (e.g., CA125, HER-2) were not included due to data availability, which could have potentially enhanced the model's predictive power.
  • Future Directions: The authors plan to implement this preoperative prediction model to guide surgical procedures and select high-risk candidates for laparoscopy. Further research could explore incorporating more advanced imaging techniques or integrating genomic data with radiomics for even more precise predictions.

Summary updated

Abstract

<b>Objective:</b> The aim of this study is to evaluate whether radiomics imaging signatures based on computed tomography (CT) could predict peritoneal metastasis (PM) in gastric cancer (GC) and to develop a nomogram for preoperative prediction of PM status. <b>Methods:</b> We collected CT images of pathological T4 gastric cancer in 955 consecutive patients of two cancer centers to analyze the radiomics features retrospectively and then developed and validated the prediction model built from 292 quantitative image features in the training cohort and two validation cohorts. Lasso regression model was applied for selecting feature and constructing radiomics signature. Predicting model was developed by multivariable logistic regression analysis. Radiomics nomogram was developed by the incorporation of radiomics signature and clinical <i>T</i> and <i>N</i> stage. Calibration, discrimination, and clinical usefulness were used to evaluate the performance of the nomogram. <b>Results:</b> In training and validation cohorts, PM status was associated with the radiomics signature significantly. It was found that the radiomics signature was an independent predictor for peritoneal metastasis in multivariable logistic analysis. For training and internal and external validation cohorts, the area under the receiver operating characteristic curves (AUCs) of radiomics signature for predicting PM were 0.751 (95%CI, 0.703-0.799), 0.802 (95%CI, 0.691-0.912), and 0.745 (95%CI, 0.683-0.806), respectively. Furthermore, for training and internal and external validation cohorts, the AUCs of radiomics nomogram for predicting PM were 0.792 (95%CI, 0.748-0.836), 0.870 (95%CI, 0.795-0.946), and 0.815 (95%CI, 0.763-0.867), respectively. <b>Conclusions:</b> CT-based radiomics signature could predict peritoneal metastasis, and the radiomics nomogram can make a meaningful contribution for predicting PM status in GC patient preoperatively.
BibTeX
@inproceedings{huang2020radiomics,
  author = {Weicai Huang and Kangneng Zhou and Yuming Jiang and Chuanli Chen and Qingyu Yuan and Zhen Han and Jingjing Xie and Shitong Yu and Zepang Sun and Yanfeng Hu and Jiang Yu and Hao Liu and Ruoxiu Xiao and Yikai Xu and Zhiwei Zhou and Guoxin Li},
  title = {{Radiomics Nomogram for Prediction of Peritoneal Metastasis in Patients With Gastric Cancer}},
  booktitle = {Frontiers in Oncology},
  year = {2020},
  doi = {10.3389/fonc.2020.01416},
}
Figure