CT-Radiomics Framework for Diagnosing Lung Cancer Linked to Cystic Airspaces

Overview

This multicenter study developed and validated a CT-based radiomics model to differentiate lung cancer associated with cystic airspaces (LCCA) from benign lesions. The model combined radiomics features from tumor and peritumor regions with clinical and radiologic data, demonstrating promising diagnostic accuracy.

Background

Lung cancer associated with cystic airspaces (LCCA) is a rare subtype of lung malignancy often misdiagnosed due to its cystic morphology, which was historically considered benign. Accurate diagnosis is challenging because cystic structures complicate tissue biopsy and increase procedural risks. Radiomics, which extracts quantitative imaging features, has shown potential in lung cancer diagnosis but has not been specifically applied to cystic lung lesions. This study aimed to establish a non-invasive radiomics model using preoperative CT images and clinical features to improve differentiation between malignant and benign cystic lung lesions.

Data Highlights

The study included patients from two centers, divided into training, internal test, and external test cohorts. Radiomics features were extracted from both tumor and 5-mm peritumoral regions on CT images, totaling 1888 features per lesion. Clinical and radiologic features such as lesion diameter, location, ground-glass opacity, spicule sign, and pleural indentation were also assessed. Feature selection involved intraclass correlation coefficient (ICC) filtering and further statistical methods to identify robust predictors for model development.

Key Findings

• The radiomics model integrating tumor and peritumor features achieved high diagnostic accuracy in distinguishing LCCA from benign cystic lesions.
• Inclusion of clinical and radiologic features improved model performance compared to radiomics alone.
• The model was validated internally and externally, demonstrating good generalizability across different patient cohorts.
• Manual segmentation of lesions and peritumoral areas was reproducible with high inter- and intra-reader agreement.
• The 5-mm peritumoral region was effective in capturing relevant imaging features associated with malignancy.

Clinical Implications

This radiomics framework offers a non-invasive, objective tool to aid clinicians in diagnosing lung cancer linked to cystic airspaces, potentially reducing reliance on risky biopsies. Incorporating both tumor and peritumoral imaging features alongside clinical data enhances diagnostic confidence. Adoption of such models could improve early detection and appropriate management of LCCA.

Conclusion

The study successfully established a CT-radiomics model that accurately differentiates malignant from benign cystic lung lesions, providing a valuable adjunct to conventional imaging and clinical assessment. This approach may facilitate earlier and safer diagnosis of LCCA in clinical practice.

References

1. Fleischner Society 1996 -- Definition of lung cystic airspace
2. Lung cancer screening observational study -- 22% missed carcinomas with cystic airspaces
3. Radiomics in NSCLC diagnosis and prognosis -- Prior studies
4. Radiomics and AI in cystic lesion diagnosis -- Other organ studies
5. Vacuole sign in lung carcinoma -- Predictive value