Related Experiment Video
Updated: Jul 11, 2025
08:05
Lung CT Segmentation to Identify Consolidations and Ground Glass Areas for Quantitative Assesment of SARS-CoV Pneumonia
Published on: December 19, 2020
14.2K
CT-Based Deep-Learning Model for Spread-Through-Air-Spaces Prediction in Ground Glass-Predominant Lung Adenocarcinoma
Mong-Wei Lin1, Li-Wei Chen2, Shun-Mao Yang3
1Department of Surgery, National Taiwan University Hospital and National Taiwan University College of Medicine, Taipei, Taiwan.
Annals of Surgical Oncology
|November 13, 2023
Summary
A new deep learning model accurately predicts tumor spread through air spaces (STAS) in early lung cancer. This AI tool aids surgical planning by identifying STAS before surgery, improving patient outcomes.
Area of Science:
- Oncology
- Radiology
- Artificial Intelligence
Background:
- Sublobar resection is linked to poor prognosis in early-stage lung adenocarcinoma, particularly with tumor spread through air spaces (STAS).
- Accurate preoperative prediction of STAS is crucial for effective surgical planning in lung adenocarcinoma.
- This study focuses on early-stage lung adenocarcinoma with tumors smaller than 3 cm and a consolidation-to-tumor (C/T) ratio less than 0.5.
Purpose of the Study:
- To develop and validate a deep-learning model for the preoperative prediction of STAS in lung adenocarcinoma.
- To assess the performance of the STAS deep-learning (STAS-DL) model against other prediction methods and human experts.
Main Methods:
- Retrospective enrollment of 581 lung adenocarcinoma patients from two institutions (2015-2019).
- Development of the STAS-DL model incorporating solid components gated (SCG) feature extraction for STAS prediction.
- External validation and comparison of STAS-DL against STAS-DL without SCG (STAS-DLwoSCG), radiomics model, C/T ratio, and thoracic surgeons using AUC, accuracy, and decision curve analysis.
Main Results:
- The STAS-DL model achieved the highest performance in the testing set (123 patients) with an AUC of 0.82 and accuracy of 74%.
- STAS-DL significantly outperformed STAS-DLwoSCG (70% accuracy) and thoracic surgeons (AUC 0.68).
- The STAS-DL model demonstrated the highest standardized net benefit, indicating superior clinical utility.
Conclusions:
- The developed STAS-DL model shows significant potential for accurate preoperative STAS prediction in lung adenocarcinoma.
- This AI-driven approach can aid surgical decision-making for early-stage, ground glass-predominant lung adenocarcinoma.
- The STAS-DL model offers a promising tool to improve surgical planning and patient outcomes.
