JoVE - Journal of Visualized Experiments
JoVE Visualize
Contact Us

A radiomics-based machine learning model and SHAP for predicting spread through air spaces and its prognostic implications in stage I lung adenocarcinoma: a multicenter cohort study

  • 0Department of Thoracic Surgery, Tianjin Chest Hospital, No.261, Taierzhuang South Road, Jinnan District, Tianjin, China.
Cancer Imaging : the Official Publication of the International Cancer Imaging Society +

|

September 30, 2025

|

September 30, 2025

View abstract on PubMed

Summary

This summary is machine-generated.

A new machine learning model accurately predicts tumor spread in early-stage lung adenocarcinoma using CT imaging. This radiomics approach offers a promising tool for personalized treatment strategies, outperforming traditional clinical models.

Area Of Science

  • Radiomics and Machine Learning in Oncology
  • Quantitative Imaging Analysis
  • Thoracic Surgery and Oncology

Background

  • Postoperative recurrence in lung adenocarcinoma is high, especially with tumor spread through air spaces.
  • Early-stage lung adenocarcinoma necessitates reliable preoperative prediction models for treatment adjustment.
  • Current detection and resection methods have limitations in predicting recurrence risk.

Purpose Of The Study

  • To develop and validate a preoperative prediction model for tumor spread through air spaces in stage I lung adenocarcinoma.
  • To compare the performance of a radiomics-based model with traditional clinical models.
  • To integrate quantitative imaging features with clinical data for improved prediction accuracy.

Main Methods

  • A multicenter retrospective study involving 609 patients with pathological stage I lung adenocarcinoma.
  • Extraction and filtering of quantitative imaging features from CT scans of the primary tumor and peritumoral regions.
  • Development of radiomics, clinical, and combined prediction models using machine learning, including elastic net regression.

Main Results

  • The radiomics model demonstrated high accuracy (AUC up to 0.829) in predicting tumor spread.
  • The combined model integrating imaging and clinical features achieved the highest performance (AUC up to 0.894).
  • Radiomics model significantly outperformed the clinical model (AUC 0.807 vs. 0.689).
  • Predicted tumor spread correlated with significantly shorter progression-free survival.

Conclusions

  • A novel machine learning model integrating radiomics features from tumor and peritumoral regions can preoperatively predict tumor spread in stage I lung adenocarcinoma.
  • The developed model shows superior performance compared to traditional clinical models.
  • Quantitative imaging analysis holds significant potential for personalizing lung cancer treatment strategies.

Keywords: