Machine learning-based model for prediction of KCNJ5 mutation in unilateral primary aldosteronism based on [68Ga]Ga-Pentixafor PET/CT

BACKGROUND: KCNJ5 mutations enhance aldosterone synthase expression and are closely associated with the prognosis of unilateral primary aldosteronism (UPA). This study developed machine learning-based models with [68Ga]Ga-Pentixafor PET/CT for predicting KCNJ5 mutations in patients with UPA. RESULTS: Most of the clinical characteristics and all of [68Ga]Ga-Pentixafor PET/CT parameters differed significantly between the KCNJ5-MT and KCNJ5-WT patients. Among these three models based on 40 min LCR, ΔLCR, size, body mass index, creatinine, duration of hypokalemia, serum renin, age, preoperative defined daily dose, the XGBoost model had the highest predictive efficacy in the training group and the AUC was 0.915; the AUC of the AdaBoost and RF was 0.914, 0.911, respectively. The AdaBoost model had the highest predictive efficacy in the test group and the AUC was 0.866; the AUC of the XGBoost and RF was 0.844, 0.859, respectively, however, there was no significant difference in diagnostic performance between the three models. Patients with KCNJ5-MT exhibit different general clinical characteristics to those with the KCNJ5-WT, but there is little difference in the initial surgical outcome assessment. CONCLUSIONS: This machine learning models based on [68Ga]Ga-Pentixafor PET/CT may achieve promising diagnostic efficacy for predicting KCNJ5 mutations. [68Ga]Ga-Pentixafor PET/CT parameters are the main predictors for KCNJ5 mutations, suggests its potential as a noninvasive imaging biomarker.