Predicting High-risk Lung Adenocarcinoma in Solid and Part-solid Nodules on Low-dose CT: A Multicenter Study.

Area of Science:

• Radiology and Oncology
• Medical Imaging Analysis
• Artificial Intelligence in Medicine

Background:

• High-risk features in lung adenocarcinomas (LUADs) like visceral pleural invasion and lymph node metastasis are linked to poor prognosis.
• Accurate identification of these high-risk LUADs is crucial for timely and effective patient management.

Purpose of the Study:

• To develop and validate a radiomic model and a radiographic model using low-dose CT (LDCT) for predicting high-risk LUADs.
• To compare the diagnostic performance of the radiomic and radiographic models in identifying high-risk LUADs within solid and part-solid nodules.

Main Methods:

• Retrospective analysis of 658 pathologically confirmed LUADs across four centers.
• Construction of radiomic and radiographic models using multivariable logistic regression based on extracted features.
• Validation of model performance using internal and external datasets, assessed by Area Under the Receiver Operating Characteristic Curve (AUC).

Main Results:

• The radiomic model incorporated three key features: GLCM_Correlation, GLSZM_SmallAreaEmphasis, and GLDM_LargeDependenceHighGrayLevelEmphasis.
• The radiographic model included maximal diameter, consolidation/tumor ratio (CTR), spiculation, and pleural indentation.
• The radiomic model demonstrated superior diagnostic performance with AUCs of 0.916 (internal) and 0.938 (external), significantly outperforming the radiographic model (0.868 and 0.880, respectively).

Conclusions:

• The developed LDCT-based radiomic model effectively identifies high-risk LUADs in solid and part-solid nodules.
• This non-invasive radiomic model shows promising diagnostic performance.
• The model may support personalized decision-making within lung cancer screening programs.