Virtual biopsy through CT imaging: can radiomics differentiate between subtypes of non-small cell lung cancer?

Objective This study evaluated the performance of CT radiomics in distinguishing between lung adenocarcinoma (ADC) and squamous cell carcinoma (SCC) at baseline imaging, exploring its potential as a noninvasive virtual biopsy. Materials and methods A retrospective analysis was conducted, enrolling 330 patients between September 2015 and January 2023. Inclusion criteria were histologically proven ADC or SCC and baseline contrast-enhanced chest CT. Exclusion criteria included significant motion artifacts and nodules < 6 mm. Radiological features, including lung lobe affected, peripheral/central location, presence of emphysema, and T/N radiological stage, were assessed for each patient. Volumetric segmentation of lung cancers was performed on baseline CT scans at the portal-venous phase using 3DSlicer software (v5.2.2). A total of 107 radiomic features were extracted and selected using the least absolute shrinkage and selection operator (LASSO) and tenfold cross-validation. Multivariable logistic regression analysis was employed to develop three predictive models: radiological features-only, radiomics-only, and a combined model, with statistical significance set at p < 0.05. Additionally, an independent external validation cohort of 16 patients, meeting the same inclusion and exclusion criteria, was identified. Results The final cohort comprised 200 ADC and 100 SCC patients (mean age 68 +/- 10 years, 184 men). Two radiological and 21 radiomic features were selected (p < 0.001). The Radiological model achieved AUC 0.73 (95% CI 0.68-0.78, p < 0.001), 72.3% accuracy. The radiomics model achieved AUC 0.80 (95% CI 0.75-0.85, p < 0.001), 75.6% accuracy. The combined model achieved AUC 0.84 (95% CI 0.80-0.88, p < 0.001), 75.3% accuracy. External validation (n = 15) yielded AUC 0.78 (p = 0.05). Conclusion The combined radiologic-radiomics model showed the best performance in differentiating ADC from SCC.