Related Experiment Video
Updated: Jul 15, 2026
Comparison of Predictive Performance of Three Lymph Node Staging Systems in Colorectal Signet Ring Cell Carcinoma Based on Machine Learning Model
Published on: April 18, 2025
Accuracy of machine learning in detecting lymph node metastasis of esophageal cancer: a systematic review and
Jing Chen1, Jiren Weng1, Xuefeng Lin1
1Affiliated Yueqing Hospital, Wenzhou Medical University, Wenzhou, China.
Background:
Currently, the detection of early lymph node metastasis in esophageal cancer during preoperative assessments is still challenging. Machine learning has been used in the detection of early lymph node metastasis in esophageal cancer. However, its detection accuracy remains controversial. Therefore, this systematic review and meta-analysis aimed to explore the accuracy of machine learning in the detection of lymph node metastasis in esophageal cancer.
Methods:
PubMed, Embase, Cochrane, and Web of Science databases were searched for related studies published before May 7, 2026. Studies were excluded based on the following criteria: meta-analyses, reviews, guidelines, and expert opinions; studies solely conducting risk factor analysis without constructing complete machine learning models; studies failing to report essential model accuracy evaluation metrics; and studies merely validating mature scales without machine learning model development. The PROBAST tool was used to assess the risk of bias in the included studies. Subgroup analyses were performed according to various datasets and modeling variables, including explainable clinical features, genomics, radiomics, and the combination of radiomics with clinical features.
Results:
In total, 49 original studies were included, of which 20 used radiomics, including 19,755 medical records. The meta-analysis revealed that in the validation dataset, the C-index of machine learning was 0.79 [95% confidence interval (CI): 0.76-0.83, I2=43.9%, P=0.058] in detecting lymph node metastasis of esophageal cancer. Subgroup analysis revealed that the C-index was 0.82 (95% CI: 0.80-0.84, I2=50.1%, P=0.01) for machine learning models based on both radiomics and clinical features.
Conclusions:
Given clinical applicability, cost, and detection accuracy, machine learning models based on both radiomics and clinical features, with a higher C-index, appear to be a more favorable technique for detecting the early lymph node metastasis status of esophageal cancer compared with machine learning models based on clinical features alone. However, given its inherent heterogeneity, the results should be interpreted with caution and need to be validated in future research.
