Evaluation of a fusion model combining deep learning models based on enhanced CT images with radiological and clinical features in distinguishing lipid-poor adrenal adenoma from metastatic lesions

Abstract

Objective: To evaluate the diagnostic performance of a machine learning model combining deep learning models based on enhanced CT images with radiological and clinical features in differentiating lipid-poor adrenal adenomas from metastatic tumors, and to explain the model's prediction results through SHAP(Shapley Additive Explanations) analysis.

Methods: A retrospective analysis was conducted on abdominal contrast-enhanced CT images and clinical data from 416 pathologically confirmed adrenal tumor patients at our hospital from July 2019 to December 2024. Patients were randomly divided into training and testing sets in a 7:3 ratio. Six convolutional neural network (CNN)-based deep learning models were employed, and the model with the highest diagnostic performance was selected based on the area under curve(AUC) of the ROC. Subsequently, multiple machine learning models incorporating clinical and radiological features were developed and evaluated using various indicators and AUC.The best-performing machine learning model was further analyzed using SHAP plots to enhance interpretability and quantify feature contributions.

Results: All six deep learning models demonstrated excellent diagnostic performance, with AUC values exceeding 0.8, among which ResNet50 achieved the highest AUC. Among the 10 machine learning models incorporating clinical and imaging features, the extreme gradient boosting(XGBoost) model exhibited the best accuracy(ACC), sensitivity, and AUC, indicating superior diagnostic performance.SHAP analysis revealed contributions from ResNet50, RPW, age, and other key features in model predictions.

Conclusion: Machine learning models based on contrast-enhanced CT combined with clinical and imaging features exhibit outstanding diagnostic performance in differentiating lipid-poor adrenal adenomas from metastases.

Keywords: Adenoma; Adrenal gland; Automatic segmentation; Deep learning; Machine learning; Metastasis.

Fig. 1

The arterial phase CT contrast-enhanced images, after preprocessing and automatic segmentation, are fed into the deep learning model. The deep learning features generated as output are then combined with clinical features to establish a machine learning model. The SHAP method is applied to provide a visual explanation of the model and to assess the predictive value of each factor for the final outcome

Fig. 2

A and B represent the trends of model Loss and ACC as the number of iterations increases, respectively

Fig. 3

ROC curves and corresponding AUC values of six deep learning models for the differentiation of adrenal adenomas and metastases

Fig. 4

The ROC curves of the 10 machine learning models in the training and testing sets; A: ROC curves for the training set; B: ROC curves for the test set

Fig. 5

SHAP analysis is used for model interpretation. A: SHAP summary plot, which provides an overall view of the impact of each feature on model output, with red indicating higher values of each variable and blue indicating lower values. B: Dependence plot, which visualizes the correlation between important features. C and D: SHAP waterfall plots for two cases, offering local explanations for model predictions on individual instances; C is for a negative case (adrenal adenoma), and D is for a positive case (adrenal metastatic tumor)