Back to 2025 Resident Posters

---

Anderson Sarcoma Reconstruction Outcomes CALculator (SaRCal):A Validated, Radiomics-based, Machine Learning-Powered Tool
Rami Elmorsi, MD¹, Luis Camacho, MD², David D. Krijgh, MD³, Heather Lyu, MD, MBI¹, Margaret S. Roubaud, MD¹, Keila Torres, MD, PhD¹, Valerae O. Lewis, MD¹, Christina L. Roland, MD¹, Alexander F. Mericli, MD¹.
¹MD Anderson Cancer Centre, Houston, TX, USA, ²Stanford University, San Francisco, CA, USA, ³University Medical Center Utrecht, Utech, Netherlands.

Introduction: Limb-sparing soft tissue sarcoma (STS) resections pose complex reconstructive challenges. We introduce a novel, validated, machine-learning (ML) powered tool to guide reconstructive decision making.
Methods: A 6-year (2016-2021) clinicoradiomic study of limb-sparing STS established the largest database of its kind, which was in-turn leveraged to construct SaRCal. Variables imputed into the ML models represented clinical and radiographic factors present in the preoperative setting. Of 316 cases, 269 (85%) were used to tune, train, and validate ML models (Lasso Regression, Gaussian Bayes, and FasterRisk) with 50-fold random sampling, leave-one-out, and 10-fold cross-validations of internal validity. The remaining 47 (15%) were reserved to assess external validity.
Results: Lasso regression learners excelled in predicting minor complications (no re-intervention), major complications (requiring re-intervention), infections, wound dehiscence, and seromas, with AUCs, accuracies, and precisions of 79-92%, 72-94%, and 73-94% on external validation(Table). Best-performing models per outcome was converted into a nomographic calculator, constituting 1-of-5 segments of SaRCal (Figure 1 and 2).
Conclusion: SaRCal is a highly-predictive ML-powered tool that can be employed preoperatively to enhance counseling and patient-specific reconstructive planning.

	Area under ROC Curve	Classification Accuracy	F1-Score	Precision	Recall
Minor complications	0.79	0.72	0.73	0.73	0.72
Major complications	0.81	0.87	0.87	0.87	0.87
Surgical site infections	0.92	0.89	0.90	0.90	0.89
Wound dehiscence	0.89	0.89	0.88	0.89	0.89
Seroma	0.83	0.94	0.94	0.94	0.94

---

Back to 2025 Resident Posters