Radiomics Signature on Computed Tomography Imaging: Association With Lymph Node Metastasis in Patients With Gastric Cancer

Yuming Jiang, Wei Wang, Chuanli Chen, Xiaodong Zhang, Xuefan Zha, Wenbing Lv, Jingjing Xie, Weicai Huang, Zepang Sun, Yanfeng Hu, Jiang Yu, Tuanjie Li, Zhiwei Zhou, Yikai Xu, Guoxin LiFrontiers in Oncology 2019 · DOI

TL;DR

This study introduces a novel radiomics signature derived from CT scans that accurately predicts lymph node (LN) metastasis in gastric cancer (GC) patients. By integrating this signature with clinical factors into a nomogram, the researchers developed a powerful tool for preoperative, individualized prediction of LN status, potentially improving treatment decisions.

What Problem Does This Paper Address?

Accurately determining the extent of lymph node (LN) metastasis in gastric cancer (GC) is crucial for guiding treatment decisions and predicting patient prognosis. Current methods for assessing LN status preoperatively, such as CT scans, often lack sufficient accuracy. While some clinicopathological factors can offer clues, they are often only available after surgery. This leaves a critical gap in preoperative decision-making, particularly for early-stage GC where less invasive treatments might be considered, or for advanced cases where neoadjuvant chemotherapy could be beneficial. The paper aims to bridge this gap by developing a more precise, non-invasive method for predicting LN metastasis using radiomics.

Methods

The researchers employed a retrospective approach, analyzing CT images from a large cohort of 1,689 GC patients across three centers. They extracted a multitude of quantitative radiomic features from the CT scans using specialized texture analysis software. To identify the most predictive features, they utilized a Least Absolute Shrinkage and Selection Operator (LASSO) logistic regression model. This technique is effective for feature selection in high-dimensional datasets, helping to avoid overfitting.

A radiomics signature, termed the "Rad-score," was then constructed based on these selected features. This signature was used to build a prediction model, which was further refined by incorporating established clinicopathological variables (like differentiation status, CA199 levels, clinical T and N stages). The final output was a radiomics nomogram, a graphical tool designed for easy, individualized preoperative prediction of LN status. The performance of this nomogram was rigorously evaluated using calibration, discrimination (measured by C-index and AUC), and clinical usefulness (assessed by decision curve analysis).

Datasets

The study utilized three independent cohorts of gastric cancer patients:

  • Training cohort: 312 patients from Nanfang Hospital of Southern Medical University.
  • Internal validation cohort: 360 patients from Nanfang Hospital of Southern Medical University.
  • External validation cohort: 1,017 patients from Sun Yat-sen University Cancer Center and the third affiliated hospital of Southern Medical University.

These datasets were used to train the radiomics signature and nomogram, and then to validate their predictive performance.

Key Insights and Ideas

  • Radiomics as a powerful predictor: The study demonstrates that quantitative features extracted from CT images (radiomics) can capture information about LN metastasis that is not readily apparent to the human eye. This suggests that CT scans contain a wealth of untapped data for cancer prognostication.
  • Feature selection is key: The use of LASSO regression was crucial in distilling a large number of radiomic features down to a concise set that effectively predicts LN status, preventing model complexity and potential overfitting.
  • Integration is superior: Combining the radiomics signature with traditional clinicopathological factors in a nomogram significantly improved predictive accuracy compared to using either component alone. This highlights the synergistic power of integrating imaging-derived features with clinical information.
  • Clinical utility: The developed nomogram is presented as an "easy-to-use" tool, suggesting its potential for direct application in clinical practice to aid in preoperative decision-making.

Key Results

The radiomics signature (Rad-score) was found to be a significant and independent predictor of pathological LN stage across all three cohorts (training, internal validation, and external validation).

  • The radiomics nomogram demonstrated good discrimination and calibration.
  • C-index values for the nomogram ranged from 0.788 (training) to 0.829 (external validation), indicating strong predictive ability.
  • AUC values for predicting pN1, pN2, and pN3 stages versus pN0 were consistently high, for example, reaching 0.949 for pN3 vs. pN0 in the training cohort.
  • Decision curve analysis showed that the radiomics nomogram provided greater clinical benefit than using clinical N stage alone or treating all patients, particularly when the threshold probability for intervention was above 10%.

Limitations and Future Work

The authors acknowledge several limitations:

  • Multi-center data from different scanners: The study used CT images acquired from different scanners, which can introduce variability in radiomic features. While they attempted to mitigate this, it remains a potential confounder.
  • Contrast agent impact: The use of contrast agents might influence radiomic features, and normalization may not fully overcome this.
  • Retrospective nature: The study was retrospective, and a prospective, multi-center study is needed for further validation.
  • Limited scope of predictors: While the nomogram incorporated key clinicopathological factors, other predictive biomarkers could potentially improve its accuracy.

Future work suggested includes:

  • Radiogenomics: Exploring the link between radiomics features and underlying tumor genomics to develop even more precise predictive models.
  • Integration with SN mapping: Combining radiomics with sentinel node (SN) mapping techniques for improved intraoperative guidance and avoidance of unnecessary extensive surgery.
  • Prospective validation: Conducting larger, prospective studies to confirm the findings and refine the nomogram.

Summary updated

Abstract

<b>Background:</b> To evaluate whether radiomic feature-based computed tomography (CT) imaging signatures allow prediction of lymph node (LN) metastasis in gastric cancer (GC) and to develop a preoperative nomogram for predicting LN status. <b>Methods:</b> We retrospectively analyzed radiomics features of CT images in 1,689 consecutive patients from three cancer centers. The prediction model was developed in the training cohort and validated in internal and external validation cohorts. Lasso regression model was utilized to select features and build radiomics signature. Multivariable logistic regression analysis was utilized to develop the model. We integrated the radiomics signature, clinical T and N stage, and other independent clinicopathologic variables, and this was presented as a radiomics nomogram. The performance of the nomogram was assessed with calibration, discrimination, and clinical usefulness. <b>Results:</b> The radiomics signature was significantly associated with pathological LN stage in training and validation cohorts. Multivariable logistic analysis found the radiomics signature was an independent predictor of LN metastasis. The nomogram showed good discrimination and calibration. <b>Conclusions:</b> The newly developed radiomic signature was a powerful predictor of LN metastasis and the radiomics nomogram could facilitate the preoperative individualized prediction of LN status.
BibTeX
@inproceedings{jiang2019radiomics,
  author = {Yuming Jiang and Wei Wang and Chuanli Chen and Xiaodong Zhang and Xuefan Zha and Wenbing Lv and Jingjing Xie and Weicai Huang and Zepang Sun and Yanfeng Hu and Jiang Yu and Tuanjie Li and Zhiwei Zhou and Yikai Xu and Guoxin Li},
  title = {{Radiomics Signature on Computed Tomography Imaging: Association With Lymph Node Metastasis in Patients With Gastric Cancer}},
  booktitle = {Frontiers in Oncology},
  year = {2019},
  doi = {10.3389/fonc.2019.00340},
}
Figure