An evidential reasoning based model for diagnosis of lymph node metastasis in gastric cancer

Zhiguo Zhou, Fang Liu, Licheng Jiao, Zhi-Long Wang, Xiao-Peng Zhang, Xiao-Dong Wang, Xiao-Zhuo Luo

Introduction

Accurate preoperative diagnosis of lymph node metastasis (LNM) is a cornerstone of effective treatment planning for gastric cancer. It dictates the extent of surgical resection and the necessity of adjuvant therapies. Traditionally, clinicians have relied on the anatomical size of lymph nodes observed via Computed Tomography (CT) or Magnetic Resonance Imaging (MRI) as the primary indicator of malignancy. However, size-based criteria are notoriously unreliable; inflammatory responses can cause benign nodes to swell, while metastatic deposits can exist in nodes of normal size. This diagnostic "noise" leads to suboptimal sensitivity and specificity.

To address these limitations, researchers have developed a novel diagnostic framework based on Evidential Reasoning (ER). By integrating multiple qualitative and quantitative CT-scan indicators, this model moves beyond the "size paradigm" to capture the broader biological behavior of the primary tumor and its surrounding environment.

Methodology

The proposed model utilizes the Dempster-Shafer theory of evidence, specifically the Evidential Reasoning approach, which is adept at handling uncertainty and fusing heterogeneous data types.

Indicator Selection and Extraction

The study utilized a dataset of 175 consecutive patients from the Peking University Cancer Hospital & Institute. For each patient, radiologists manually extracted eight distinct indicators from preoperative CT scans, including:

  • Serosal Invasion: Assessment of tumor penetration through the stomach wall.
  • Tumor Enhancement Pattern: How the tumor absorbs contrast agent.
  • Tumor Thickness: Quantitative measurement of the primary lesion.
  • Lymph Node Size: The traditional metric (short-axis diameter).
  • Bormann Classification: Morphological type of the gastric cancer.

Model Development

The ER algorithm standardizes these diverse indicators and assigns them weights determined through an $l_2$ minimization optimization process. Unlike standard regression models, ER calculates a "residual" for potential diagnostic classes (Metastasis vs. No Metastasis). The diagnosis is determined by the class with the minimum residual, effectively "fusing" the evidence from all eight indicators into a single decision.

Results and Performance Comparison

The ER model was validated using 5-fold cross-validation against post-operative histopathology—the gold standard for LNM diagnosis.

Superiority Over Size-Based Metrics

The most striking result was the model's improvement over traditional radiologist assessments. Using a standard 7.7mm size cutoff, radiologists achieved a sensitivity of approximately 63.4%. In contrast, the ER model achieved a sensitivity of 89.51% and a specificity of 80%.

ROC curve comparing the ER Model against the traditional Lymph Node (LN) Size metric. The ER model shows a significantly larger Area Under the Curve (AUC), indicating superior diagnostic accuracy.

As shown in the ROC analysis above, the ER model (AUC 0.829) significantly outperforms the standalone LN size metric (AUC 0.757). This suggests that the inclusion of biological indicators like serosal invasion provides critical context that size alone lacks.

Benchmarking Against Machine Learning

The researchers also compared the ER approach to several established machine learning algorithms, including Artificial Neural Networks (ANN), Support Vector Machines (SVM), and Logistic Regression.

Comparison of ROC curves for various diagnostic models. While several machine learning techniques perform well, the ER model remains highly competitive and robust across testing datasets.

While some models like the Gaussian-kernel SVM showed high performance on training data, the ER model demonstrated superior robustness and reliability during cross-validation. It generally outperformed ANN (AUC 0.791) and Logistic Regression (AUC 0.793), proving that the evidential reasoning framework is particularly well-suited for the "fuzzy" nature of clinical imaging data.

Discussion and Future Directions

The success of the ER model highlights a shift toward "biological imaging." By treating LNM as a consequence of the primary tumor's aggressiveness rather than just a localized physical swelling, the model provides a more nuanced and accurate diagnostic tool.

However, several hurdles remain before widespread clinical adoption. The current reliance on manual feature extraction by radiologists is a bottleneck; future work could involve using deep learning to automate this process. Additionally, the model requires validation against larger, multi-center datasets to ensure its weights remain valid across different patient populations and imaging hardware.

In conclusion, the evidential reasoning approach offers a powerful method for multi-factor data fusion in oncology, significantly improving the preoperative detection of lymph node metastasis in gastric cancer patients.

Summary updated

Abstract

Lymph node metastasis (LNM) in gastric cancer is a very important prognostic factor affecting long-term survival. Currently, several common imaging techniques are used to evaluate the lymph node status. However, they are incapable of achieving both high sensitivity and specificity simultaneously. In order to deal with this complex issue, a new evidential reasoning (ER) based model is proposed to support diagnosis of LNM in gastric cancer. There are 175 consecutive patients who went through multidetector computed tomography (MDCT) consecutively before the surgery. Eight indicators, which are serosal invasion, tumor classification, tumor enhancement pattern, tumor thickness, number of lymph nodes, maximum lymph node size, lymph node station and lymph node enhancement are utilized to evaluate the tumor and lymph node through CT images. All of the above indicators reflect the biological behavior of gastric cancer. An ER based model is constructed by taking the above indicators as input index. The output index determines whether LNM occurs for the patients, which is decided by the surgery and histopathology. A technique called k-fold cross-validation is used for training and testing the new model. The diagnostic capability of LNM is evaluated by receiver operating characteristic (ROC) curves. A Radiologist classifies LNM by adopting lymph node size for comparison. 134 out of 175 cases are cases of LNM, and the remains are not. Eight indicators have statistically significant difference between the positive and negative groups. The sensitivity, specificity and AUC of the ER based model are 88.41%, 77.57% and 0.813, respectively. However, for the radiologist evaluating LNM by maximum lymph node size, the corresponding values are only 63.4%, 75.6% and 0.757. Therefore, the proposed model can obtain better performance than the radiologist. Besides, the proposed model also outperforms other machine learning methods. According to the biological behavior information of gastric cancer, the ER based model can diagnose LNM effectively and preoperatively.
BibTeX
@article{zhou2013evidential,
  author = {Zhiguo Zhou and Fang Liu and Licheng Jiao and Zhi-Long Wang and Xiao-Peng Zhang and Xiao-Dong Wang and Xiao-Zhuo Luo},
  title = {{An evidential reasoning based model for diagnosis of lymph node metastasis in gastric cancer}},
  journal = {BMC Medical Informatics Decis. Mak.},
  year = {2013},
  doi = {10.1186/1472-6947-13-123},
}
Figure