Prediction of solitary ≤3-cm small adrenal metastasis based on ¹⁸F-FDG PET/CT using a support vector machine model in...

The purpose of the present study was to develop a simplified scoring system based on the support vector machine (SVM) classification method to improve the diagnostic performance of ¹⁸F-fluorodeoxyglucose (FDG) positron emission tomography/computed tomography (PET/CT) in differentiating solitary small (≤3 cm) adrenal metastases from benign lesions in patients with lung cancer. A total of 197 patients with histopathologically confirmed diagnoses of lung cancer were retrospectively included. All patients had solitary adrenal lesions (long diameter ≤3 cm) showing hyperattenuating features (unenhanced CT values ≥10 HU) on pre-treatment ¹⁸F-FDG PET/CT scans. The cohort included 128 cases of metastases and 69 cases of benign lesions. SVM models were developed using five adrenal lesion features and one primary lung cancer feature. Model performance was evaluated by comparing the area under the receiver operating characteristic curve (AUC) and accuracy across eight candidate models. The optimal model was further simplified based on feature weights and value distributions to enhance clinical applicability. The best-performing SVM model demonstrated maximum, minimum and mean accuracies of 98.0% (95% CI, 93.9-100%), 71.4% (95% CI, 57.1-83.7%) and 84.3% (95% CI, 69.4-91.8%), respectively, with corresponding AUC values of 1.000, 0.770 and 0.913, respectively. The simplified scoring system stratified adrenal lesions into three diagnostic categories: Scores <5, benign; scores >6.5, metastatic; and scores 5-6.5, suspicious for metastasis. In conclusion, the present study developed an SVM-based model using ¹⁸F-FDG PET/CT imaging features to differentiate solitary small adrenal metastases in patients with lung cancer. The maximum standardized uptake value of the adrenal lesions was the predominant predictive feature. After simplification, the model was converted into a clinically practical scoring system, which may assist clinicians in streamlining clinical staging decisions for patients with lung cancer.