Dual-layer spectral detector computed tomography multiparameter machine learning model for prediction of lymph node metastases in esophageal squamous cell carcinoma

TL;DR

This study uses machine learning with CT scan data to predict lymph node metastasis in esophageal cancer patients, offering a non-invasive diagnostic tool.

Contribution

A novel machine learning model combining dual-layer spectral CT parameters and lymph node features for predicting metastasis in ESCC.

Findings

The logistic regression model achieved ROC-AUC values of 0.885 in training and 0.827 in testing.

Key predictors included LAD, SAD, CTV−40keV, NICV, ED, ECVV, and Nct.

The model showed strong diagnostic value with Brier scores of 0.135 and 0.172 in training and testing.

Abstract

Accurate assessment of lymph node metastasis (LNM) is essential for the staging, treatment, and prognosis of esophageal squamous cell carcinoma (ESCC). This study investigates the potential of dual-layer spectral detector computed tomography (SDCT) quantitative parameters in predicting LNM in ESCC.

The study included 158 patients with pathologically confirmed ESCC, comprising 92 patients without LNM and 66 patients with LNM. The chi-square test or Fisher’s exact test was utilized to analyze the basic clinical data and lymph node morphological features of the patients. To evaluate the differences in various SDCT quantitative parameters between the LNM and non-LNM groups, the Mann-Whitney U-test and independent sample t-test were applied. Patients were randomly assigned to training and test groups in a 7:3 ratio. The area under the receiver operating characteristic (ROC) curve (AUC) was used to assess the machine learning model. Furthermore, decision curve analysis (DCA) was performed to evaluate the model’s diagnostic efficacy, leading to the development of a nomogram.

Both univariate and LASSO analyses identified LAD, SAD, CTV−40keV, NICV, ED, ECVV, and Nct as significant predictors of LNM in EC. The logistic regression model demonstrated superior performance, achieving ROC-AUC values of 0.885 and 0.827 in the training and test cohorts, respectively. The Brier scores for the combined model were 0.135 and 0.172 in the training and test cohorts, respectively.

The logistic regression model, which integrates SDCT quantitative parameters and lymph node morphological features, exhibited substantial diagnostic value for assessing LNM in EC. It demonstrated excellent diagnostic efficacy and provides a non-invasive evaluation method for clinical practice.

Figures