The CT-based multi-omics model for prognostic prediction and tumor microenvironmental alterations in retroperitoneal liposarcoma: A retrospective multicohort analysis.

This is the authors' abstract. We don't add key points for this paper.

11564 Background: Retroperitoneal liposarcomas (RPLS) are tumours with a poor prognosis. Due to the tumour heterogeneity of RPLS, no effective prognostic index has been established. Radiomics has the potential to non-invasively characterize the so-called radiological phenotype of RPLS. We aimed to develop and independently validate a CT-based multi-omics model for predicting prognosis and tumour microenvironment (TIME) in RPLS to guide neoadjuvant therapy in clinical application. Methods: In this retrospective multicohort study, we used five independent cohorts of RPLS patients at our centre (n = 173, Zhongshan hospital, Fudan University, China), and the external validation cohort comprising RPLS patients from another medical centre (n = 184, Shanghai Public Health Clinical Center, Fudan University, China), to develop and validate a CT-based multi-omics model predictive of neoadjuvant chemotherapy (n = 10) and radiotherapy (n = 20) response by combining contrast-enhanced CT images and RNA-seq genomic data (n = 59) from tumour samples to assess the immune and metabolic landscape. Using the training dataset, a CT-based radiomics workflow was developed, including manual delineation, sub-segmentation, feature extraction, and predictive model building by LASSO, which was validated according to the area under the curve (AUC) for the internal validation set and external validation set. The clinical-radiomic model was developed through Univariate and Multivariate Cox regression analysis. Kaplan-Meier and log-rank tests were used for the survival analysis. Results: A total of 436 patients recruited between September 2009 and October 2021 were included in the training, internal validation, and external validation sets. The OS-related CT-based radiomics models in the training, internal validation, and external validation sets predicted the AUC values of postoperative 5-year OS as 0.71, 0.85, and 0.82, respectively. Strikingly, the clinical-CT radiomics model showed superior prognostic ability in predicting postoperative 5-year OS of 0.83, 0.85, and 0.86. Furthermore, the clinical-CT radiomics risk score was significantly decreased in post-neoadjuvant chemotherapy and radiotherapy with more tertiary lymphoid structures (TLS) and immune cell infiltration, such as plasma cells and B cells, compared to pretreatment. KEGG and ssGSEA analysis further indicated that retinol metabolism was most significantly inhibited in the high clinical-CT radiomics risk score group. Conclusions: Our validated clinical-CT radiomics model can predict the prognosis and TIME of RPLS with excellent performance. This may be a promising way to improve prognosis prediction and risk stratification of patients treated with neoadjuvant therapy, when validated by further prospective randomised trials.