1268P - Liquid biopsy and CT-based multi-omics fusion enhances differential diagnosis of early-stage lung adenocarcinoma

Background

Lung cancer is a leading cause of cancer-related deaths worldwide. Early detection using low-dose CT (LDCT) has shown promise in reducing mortality rates. Nevertheless, LDCT screening poses challenges, including a high false-positive rate and the risk of overdiagnosis. AI and biomarkers are crucial directions for the future of lung cancer screening. This study investigates the potential of multi-omics fusion, combining liquid biopsy biomarkers and computed tomography (CT) features, to enhance lung cancer differential diagnosis.

Methods

A total of 146 participants were enrolled, including 111 early-stage lung adenocarcinoma patients and 35 controls. High throughput extracellular vesicle long RNA (evlRNA) sequencing were performed using serum samples obtained before lung surgery. A fusion model was developed to combine image-based features and evlRNA features.

Results

Our findings suggest that evlRNA and image-based features complement each other, achieving improved performance when evlRNA features are fused with radiological features from AI, CTR, or human experts. Our methods achieve notable differential diagnosis performance; for example, combining evlRNA and radiological AI yields a 4-class AUC of 0.919 and a benign-malignant AUC of 0.948 (sensitivity: 89.1%, specificity: 94.3%). Integrating multi-omics analysis with human experts further enhances diagnostic accuracy. Additionally, we provide post-hoc explanations using Shapley values to understand feature importance in the multi-omics modeling. Table: 1268P

Conclusions

In summary, our study demonstrated that the combination of evlRNA-based and image-based features can complement each other and that the collaborative approach between multi-omics and human analysis can further enhance the performance. These findings highlight the importance of integrating multiple modalities including effective biomarkers and CT analysis with AI algorithms, into LDCT screening through multi-omics fusion.

Clinical trial identification

Editorial acknowledgement

Legal entity responsible for the study

Yanwei Zhang.

Funding

Shanghai Science and Technology Committee.

Disclosure

All authors have declared no conflicts of interest.