1267P - Enhancing pulmonary nodule diagnosis: A combinatorial model of cfDNA methylation, plasma proteins and LDCT imaging

Background

Low dosage computer tomography (LDCT) is widely used to detect early-stage lung cancer, but concerns regarding accuracy and overdiagnosis persist. To enhance LDCT diagnosis using non-invasive molecular features, we developed a combinatorial model to distinguish between malignant and benign pulmonary nodules.

Methods

In a prospective cohort of 608 participants with pulmonary nodules, we performed targeted methylation sequencing and protein level measurement using Proximity Extension Assay. Radiomics features were extracted from LDCT images of 448 participants. A machine learning classifier, incorporating a transformer model and deep neuron network models, was trained and tested, by integrating molecular and image features.

Results

A total of 368 samples (184 benign and 184 malignant) with matched sex and age were randomly selected as the training set. The remaining 81 benign and 159 malignant samples were used as the test set. The methylation-only model had an AUC of 0.805 [95% CI 0.755-0.852], the protein-only model had an AUC of 0.816 [0.768-0.860], and the radiomics-only model achieved an AUC of 0.865 [0.812-0.912]. A combination of methylation and radiomics features generated an enhanced model with an AUC of 0.884 [0.840-0.927] (sensitivity = 0.824 [0.744-0.883], specificity = 0.772 [0.641-0.865]), outperforming models based on other combinations of two features. Integrating protein markers further improved the model (AUC = 0.895 [0.845-0.934]), with both sensitivity (0.849 [0.774-0.905]) and specificity (0.842 [0.721-0.918]) showing significant enhancements. The combinatorial model performed well across sample groups with different nodule sizes, particularly for samples with nodules no larger than 10 mm.

Conclusions

This study integrated DNA methylation, protein, and radiomics features to construct a robust combinatorial model with optimal performance, providing potential clinical utility for the management of pulmonary nodules.

Clinical trial identification

Editorial acknowledgement

Legal entity responsible for the study

The authors.

Funding

Has not received any funding.

Disclosure

W. Li, Q. He, Y. Zhang: Financial Interests, Personal, Full or part-time Employment: Singlera Genomics Inc. Z. Su: Financial Interests, Personal, Full or part-time Employment: Singlera Genomics Ltd. R. Liu: Financial Interests, Personal, Officer: Singlera Genomics Inc. All other authors have declared no conflicts of interest.