Prediction of Lymphovascular Invasion in Early-Stage Lung Adenocarcinoma Using Artificial Intelligence-Based...

Area of Science:

• Oncology
• Radiology
• Artificial Intelligence
• Biomarkers

Background:

• Lymphovascular invasion (LVI) is a critical prognostic factor in early-stage lung adenocarcinoma, associated with poorer survival outcomes.
• Accurate prediction of LVI is essential for guiding treatment decisions and improving patient management in early lung cancer.

Purpose of the Study:

• To evaluate the effectiveness of artificial intelligence (AI)-based radiomics analysis of computed tomography (CT) images for predicting lymphovascular invasion (LVI) in stage 0-I lung adenocarcinoma.
• To investigate the added value of combining radiomics data with serum microRNA (miR)-30d levels for enhancing LVI prediction.

Main Methods:

• AI-based radiomics using a modified U-Net for segmentation and a VGG-16 convolutional neural network for LVI prediction on CT images from 1265 patients.
• Development of a risk score (RS) based on 35 imaging features in a derivation cohort (840 patients) and validation in an independent cohort (425 patients).
• Analysis of extracellular vesicle (EV)-derived miR-30d levels in 47 cases to assess the integrated approach's predictive performance.

Main Results:

• AI-based radiomics achieved high accuracy in predicting LVI, with the risk score demonstrating an area under the curve of 0.899 in the derivation cohort.
• The risk score showed good performance in both derivation (sensitivity 84.8%, specificity 83.7%, accuracy 83.9%) and validation (sensitivity 82.3%, specificity 79.4%, accuracy 80.5%) groups.
• The integrated approach combining radiomics with EV-derived miR-30d levels further improved LVI prediction (sensitivity 93.3%, specificity 70.5%, accuracy 85.1%).

Conclusions:

• AI-based radiomics is a highly effective tool for predicting LVI in early-stage lung adenocarcinoma, with the developed risk score having significant clinical potential.
• The integration of EV-derived miR-30d levels modestly enhances the predictive accuracy of LVI when combined with radiomics data.