Sarcoma
Fabio Tirotta
Surgical Oncologist
University Hospital Birmingham NHS Foundation Trust, Birmingham, UK, England, United Kingdom
Fabio Tirotta
Surgical Oncologist
University Hospital Birmingham NHS Foundation Trust, Birmingham, UK, England, United Kingdom
Fabio Tirotta
Surgical Oncologist
University Hospital Birmingham NHS Foundation Trust, Birmingham, UK, England, United Kingdom
Anne-Rose Schut, n/a
Resident
Erasmus MC, Rotterdam, The Netherlands, United States
Demi Wemmers, n/a
MD
Erasmus MC, The Netherlands, United States
Stefan Klein, n/a
MD
Erasmus MC, Rotterdam, The Netherlands, United States
Jacob J. Visser, n/a
MD
Erasmus MC, Rotterdam, The Netherlands, United States
Dirk Grunhagen, MD PhD
Surgical Oncologist
Erasmus MC Rotterdam, United States
Arno van Leenders, n/a
MD
Erasmus MC, Rotterdam, The Netherlands, United States
David Hanff, n/a
MD
Erasmus MC, Rotterdam, The Netherlands, United States
Winan J. van Houdt, MD, PhD, MS
Surgical oncologist
Netherlands Cancer Institute, Division of Surgical Oncology, Amsterdam, The Netherlands
Amsterdam, Netherlands
Cornelis Verhoef, Prof. Dr.
Head of oncological surgery department
Erasmus Medical Center, Netherlands
Martijin P.A. Starmans, n/a
Post-doc researcher
Erasmus MC, Rotterdam, The Netherlands, United States
Targeting the solid component of the tumors has determined a significant improvement in differentiating between low- and high- grade retroperitoneal liposarcoma (RLPS). However, grading accuracy on biopsy remains variable. The aim of this study was to evaluate the accuracy of radiomics-based preoperative CT models at predicting tumor grading in patients with primary RLPS.
Methods:
Data on consecutive patients who underwent surgery for primary RPLS were retrospectively analyzed. Low-grade (grade 1) RPLS corresponded to WDLPS and high-grade (grade 2 and 3) RLPS to DDLPS. Three different radiomics-based models were devised: Model 1) based on manually scored imaging features; Model 2) based on CT scan images; and Model 3) based on both manually scored and CT scan imaging features. The performances of the radiomics models were evaluated using the Area Under the Curve (AUC) of the receiver operating characteristic curve, balanced classification accuracy (BCA), sensitivity, specificity, negative predictive value (NPV), and positive predictive value (PPV).
Results: Data were available for N=64 patients with RLPS, of whom n=30 with low-grade RLPS (WDLPS) and n=34 with high-grade RLPS (DDLPS). No significant differences in terms of age (p=0.46), sex (p=0.13), tumor location (p=0.52), tumor size (p=0.16), and tumor volume (p=0.45) were observed between the two groups. The resulting AUC, sensitivity, and specificity at predicting the tumor grading on the final histology was 0.84, 0.83, and 0.72 for Model 1; 0.87, 0.81, and 0.77 for Model 2, and 0.84, 0.81, and 0.74 for Model 3. In addition, BCA, NPV, and PPV were 0.77, 0.80, and 0.79 for Model 1; 0.79, 0.80, and 0.82 for Model 2; and 0.77, 0.79, and 0.80 for Model 3.
Conclusions:
The radiomics-based preoperative CT-models represent an accurate, objective, and non-invasive tool to predict preoperative grading in RLPS. These models may potentially be used as an effective alternative to preoperative biopsy. Further large multicenter studies are warranted to validate these findings.