JoVE - Journal of Visualized Experiments
JoVE Visualize
Contact Us

Development and Validation of a Computed Tomography-Based Model for Noninvasive Prediction of the T Stage in Gastric Cancer: Multicenter Retrospective Study

  • 0Department of General Surgery and Laboratory of Gastric Cancer, State Key Laboratory of Biotherapy/Collaborative Innovation Center of Biotherapy and Cancer Center, West China Hospital, Sichuan University, Chengdu, China.
Journal of Medical Internet Research +

|

October 9, 2024

|

October 9, 2024

View abstract on PubMed

Summary

This summary is machine-generated.

A hybrid deep learning and radiomics model accurately predicts gastric cancer (GC) T stage using CT scans. This advanced model improves upon traditional methods for better treatment planning.

Area Of Science

  • Oncology
  • Radiology
  • Artificial Intelligence

Background

  • Tumor-node-metastasis (TNM) staging is critical for gastric cancer (GC) treatment planning.
  • T staging, based on tumor depth, is a key component of TNM staging.
  • Prior research has explored deep learning and radiomics for GC prognosis, but T staging prediction remains underexplored.

Purpose Of The Study

  • To develop and evaluate a computed tomographic (CT)-based model for automatic T stage prediction in GC.
  • The model integrates radiomics and deep learning techniques.
  • The study aimed to enhance diagnostic accuracy for GC pathological staging.

Main Methods

  • Retrospective analysis of 771 GC patients from three centers.
  • Classification of GC into mild (T1-T2), moderate (T3), and severe (T4) stages.
  • Development of three models: radiomics, deep learning, and a hybrid approach combining both.

Main Results

  • The hybrid model achieved the highest overall classification accuracy (81.4%) in internal testing.
  • Deep learning (75.7%) and hybrid models outperformed the radiomics model (64.3%).
  • The hybrid model demonstrated superior performance in binary classification tasks, especially in external validation (AUC 0.972 for T1-T2 vs. T3-T4).

Conclusions

  • A hybrid model integrating radiomics and deep learning features shows significant promise for predicting GC pathological T stage.
  • This CT-based approach offers improved diagnostic accuracy compared to individual radiomics or deep learning models.
  • The findings suggest potential for enhanced clinical decision-making in GC management.

Keywords: