Preoperatively Predicting Risk Stratification for GISTs ≤2 cm by Radiomics Model: A Dual-center Study

Area of Science:

• Oncologic Imaging
• Radiology
• Medical Informatics

Background:

• Small gastrointestinal stromal tumors (SGISTs) pose diagnostic challenges due to their malignancy risk.
• Current preoperative evaluation methods for SGISTs are insufficient for accurate risk stratification.
• Radiomics, a novel image analysis technique, has not been previously applied to SGIST risk assessment.

Purpose of the Study:

• To develop and validate a CT radiomics model for preoperative risk stratification of SGISTs.
• To compare the performance of a radiomics model against a clinical model and a combined model.
• To establish a nomogram for improved clinical application in decision-making.

Main Methods:

• 133 patients with SGISTs were included, randomly assigned to training (n=93) and testing (n=40) sets.
• Radiomics features were extracted from CT images; LR-LASSO was used for dimensionality reduction.
• Clinical and combined models were developed and compared using ROC analysis and Delong tests.

Main Results:

• The clinical model (maximal tumor diameter) achieved an AUC of 0.641.
• The radiomics model, particularly using portal venous phase CT images, showed high discriminative ability (AUCs of 0.848 and 0.824 in training and testing sets).
• The combined model achieved AUCs of 0.862 (training) and 0.859 (testing), significantly outperforming the clinical model (P < 0.05).

Conclusions:

• The CT radiomics model demonstrates superior performance for preoperative risk stratification of SGISTs compared to clinical models.
• The combined model's nomogram aids in optimizing surgical resection decisions.
• Radiomics offers an effective, non-invasive tool for SGIST risk stratification, enhancing preoperative decision-making.