Estimation of pathological subtypes in subsolid lung nodules using artificial intelligence.

Area of Science:

• Pulmonary Medicine
• Radiology
• Artificial Intelligence

Background:

• Subsolid nodules (SSNs) pose diagnostic challenges in identifying pulmonary adenocarcinoma subtypes.
• Accurate differentiation between non-invasive and invasive adenocarcinoma is crucial for patient management.

Purpose of the Study:

• To evaluate the efficacy of artificial intelligence (AI) in differentiating pathological subtypes of invasive pulmonary adenocarcinomas within SSNs.
• To identify key imaging features that predict invasiveness in pulmonary adenocarcinomas.

Main Methods:

• Retrospective analysis of 120 SSNs from 110 patients.
• Automated extraction of qualitative and quantitative imaging characteristics using an AI system, followed by radiologist verification.
• Logistic regression and ROC analyses to identify independent risk factors and diagnostic performance.

Main Results:

• Major diameter and 3D CT value entropy were identified as independent risk factors for invasive adenocarcinoma.
• Optimal cut-off values were determined: 15.5 mm for major diameter and 5.17 for 3D CT value entropy.
• A combined model using these features achieved an AUC of 0.868, with 84.5% sensitivity and 80.6% specificity.

Conclusions:

• Major diameter and 3D CT value entropy are valuable indicators for distinguishing non-invasive from invasive adenocarcinoma.
• AI-driven analysis significantly enhances the performance in classifying pathological subtypes of invasive pulmonary adenocarcinomas in SSNs.