Abstract Body (Do not enter title and authors here): Research queation:
To evaluate clinical and CT morphological features versus CT texture analysis in differentiating primary pulmonary artery sarcoma (PAS) and central pulmonary embolism (PE).

Methods:
23 pathology-proven primary PAS patients and 54 pathology-proven central PE patients from September 2013 to September 2023 were continuously enrolled. Clinical features and qualitative CT pulmonary artery (CTPA) morphological features were retrospectively collected. The chi-square tests or Mann–Whitney U test was applied to find the significant features. Then the discriminative features were identified by univariate logistic analysis. The CTPA texture features were extracted using a radiomics software (Radiomics, Syngo.Via FRONTIER, version 1.4.0; Siemens). Minimum redundancy and maximum relevance (MRMR) method was used to select significant texture features. A clinical model and a texture model for distinguishing primary PAS from central PE were built using multivariate logistic regression. The performance of two models was assessed using the area under the curve (AUC).

Results:
10 from 1226 texture features selected by MRMR showed good performance in discriminating between the two groups (AUC > 0.80), of which 5 were chosen to build the texture model using multivariate logistic regression. These features were wavelet_LLL_glrlm_ShortRunHighGrayLevelEmphasis, wavelet_LHH_glszm_LargeAreaEmphasis, exponential_glszm_GrayLevelNonUniformity, original_glcm_Imc1 and wavelet_HLH_firstorder_Mean, respectively. Univariate logistic analysis was used to find 18 discriminative (p < 0.05) features, of which 4 were selected to build the clinical model using multivariate logistic regression. These features were lower limb edema, involvement of main pulmonary artery, involvement of pulmonary valve or right ventricular outflow tract and expansive growth in peripheral pulmonary artery. We evaluated these two prediction models based on CT textural features and clinical plus qualitative CTPA morphological features. These two models both showed high predictive accuracy. AUC of CT textural features model and clinical plus qualitative CTPA morphological features model were 0.97 (0.90 - 0.99) and 0.93 (0.85 - 0.98) respectively (p=0.36).

Conclusion:
We found very effective clinical and CTPA morphological features for the differential diagnosis of PAS and central PE, while CT texture analysis can additionally provide accurate, objective and quantitative diagnostic information.