Analysis of the diagnostic value of CT radiomics models in differentiating GIST and other mesenchymal tumors.

Abstract

Objective: To analyze the diagnostic value of computed tomography (CT) radiomics models in differentiating gastrointestinal stromal tumors (GIST) and other mesenchymal tumors.

Material and methods: A retrospective analysis of clinical data from 153 patients with pathologically confirmed gastrointestinal mesenchymal tumors treated in our hospital from July 2019 to March 2024 was conducted, including 107 cases of GIST, 18 cases of leiomyoma, and 28 cases of schwannoma. LASSO regression was used for feature selection. Logistic regression and Random Forest (RF) models were established based on selected features using machine learning algorithms, with the dataset divided into training (107 cases) and validation sets (46 cases) at a 7:3 ratio. The diagnostic performance of the models was evaluated using receiver operating characteristic (ROC) curves.

Results: In the training set, there were significant differences between GIST and non-GIST in terms of enhancement degree, age, maximum diameter, and tumor location distribution (P<0.05). A total of 180 radiomics features were extracted using A.K software. LASSO regression reduced the high-dimensional data to 13 radiomics features. Logistic regression and RF models were established based on these 13 features. The AUC for the Logistic regression model was 0.753 in the training set and 0.582 in the validation set, while the AUC for the RF model was 0.941 in the training set and 0.746 in the validation set. The RF model showed higher diagnostic performance than the Logistic regression model (P<0.05). Decision curve analysis showed that the net benefit of the RF model in differentiating GIST was superior to classifying all patients as either GIST or non-GIST and also superior to the Logistic regression model within a probability threshold range of 20%-90%.

Conclusion: The machine learning models based on radiomics features have good diagnostic value in predicting the pathological classification of GIST and other mesenchymal tumors, with the RF model showing superior diagnostic value compared to the Logistic regression model.