Prediction of lymphovascular invasion in esophageal squamous cell carcinoma by computed tomography-based radiomics analysis: 2D or 3D ?

Yang Li ¹, Xiaolong Gu ¹, Li Yang ¹

⁰Department of Computed Tomography and Magnetic Resonance Imaging, The Fourth Hospital of Hebei Medical University, Shijiazhuang, Hebei Province, 050011, China.

Cancer Imaging : the Official Publication of the International Cancer Imaging Society +

|

October 18, 2024

View abstract on PubMed

Summary

Whole-volume three-dimensional (3D) radiomics models show superior performance over one-slice two-dimensional (2D) models in predicting lymphovascular invasion (LVI) in esophageal squamous cell carcinoma (ESCC). The 3D approach offers higher accuracy and clinical utility for LVI status prediction.

Area Of Science

Radiology
Oncology
Medical Imaging Analysis

Background

Esophageal squamous cell carcinoma (ESCC) is a significant cause of cancer mortality.
Accurate prediction of lymphovascular invasion (LVI) status is crucial for ESCC staging and treatment planning.
Computed tomography (CT)-based radiomics offers a non-invasive method for tumor characterization.

Purpose Of The Study

To compare the predictive performance of one-slice 2D and whole-volume 3D CT-based radiomics models for LVI in ESCC.
To evaluate the clinical utility of these radiomics models in predicting LVI status.

Main Methods

Retrospective analysis of 224 ESCC patients with contrast-enhanced CT (CECT) images.
Extraction of 2D and 3D radiomics features from primary tumors using 1.0 mm slice thickness.
Feature selection using ICC, Wilcoxon rank-sum, Spearman correlation, and LASSO.
Model development using multivariate logistic stepwise regression.
Performance assessment via ROC curve analysis and clinical utility evaluation using DCA.

Main Results

753 2D and 1130 3D radiomics features were initially extracted; 7 features were selected for each model.
The 3D radiomics model demonstrated superior AUC values (0.930 training, 0.897 testing) compared to the 2D model (0.852 training, 0.851 testing).
The 3D model achieved higher accuracy (0.899 training, 0.788 testing) than the 2D model (0.728 training, 0.758 testing).
The 3D model showed higher specificity and positive predictive value, while the 2D model had higher sensitivity and negative predictive value.
Decision curve analysis indicated greater clinical utility for the 3D radiomics model.

Conclusions

Both 2D and 3D radiomics features are viable biomarkers for predicting LVI in ESCC.
Whole-volume 3D radiomics models significantly outperform one-slice 2D models in predicting LVI status in ESCC.
The 3D radiomics approach enhances diagnostic accuracy and clinical utility for LVI prediction in ESCC.

Keywords:

Computed tomography Esophageal squamous cell carcinoma Lymphovascular invasion Radiomics