Preoperative diagnosis of pulmonary sarcomatoid carcinoma based on CT findings and radiomics

AbstractBackground Pulmonary sarcomatoid carcinoma (PSC) is a rare subtype of non-small cell lung cancers (NSCLC), but differs in terms of prognosis and treatment strategies. Due to the rarity of PSC, there are few reports focus on the CT radiomics of PSC. However, the preoperative diagnosis of PSC is important and remains challenging. The aim of the study is to explore the feasibility of preoperative differentiation of PSC from other NSCLC based on CT findings and radiomics, so as to improve the accuracy of radiological diagnosis of PSC. Methods 31 patients with PSC and 56 patients with other NSCLC were retrospectively analyzed. CT findings included tumor size, tumor location, calcification, vacuole/cavity, pleural invasion, and low-attenuation area (LAA) ratio. A total of 851 radiomics features were extracted from each CT phase data, including the plain scan (PS), arterial phase (AP) and venous phase (VP). The training and testing cohorts were created in an 8:2 ratio, and the top-ranked 11 features were selected using least absolute shrinkage and selection operator (LASSO) method. Seven machine learning algorithms (DT, GBDT, LDA, LR, RF, SVM, and XGBoost) were applied for the differential diagnosis of PSC and other NSCLC. Results The median survival times of PSC and other NSCLC were 8 months (95% CI 2.123–13.877) and 34 months (95% CI 22.920–45.080), respectively. The mean tumor size of PSC (2.0-9.3 cm) and other NSCLC (2.1–9.7 cm) was 5 cm, and the difference was not statistically significant. Compared to other NSCLC, PSC had a larger LAA ratio (P < 0.001), with an optimal cutoff value of 16.6%, and a sensitivity and specificity of 0.806 and 0.732, respectively. In CT radiomics, PS data combined with logistic regression (LR) algorithm yielded the highest diagnostic efficacy, and the area under the curve (AUC), accuracy, sensitivity and specificity were 0.972, 0.944, 0.833 and 1.000, respectively. Conclusions CT findings and radiomics showed efficient performance in the differential diagnosis of PSC from other NSCLC, which is helpful for the preoperative diagnosis of PSC.