Comprehensive Summary
Separate renal function, or split renal function, is an important assessment in clinical decision making for kidney donation, nephrectomy, and malformation/ obstruction indications. This method of determining the functionality of each kidney to overall kidney function is often tedious and high cost, presenting an attractive alternative in the use of plain CT images combined with artificial intelligence (AI) methods. Specifically, the usage of deep learning-based automatic segmentation and radiomics modeling provides a novel approach to assessing separate renal function. Through a retrospective study on 281 patients, where ElasticNet algorithm was used for radiomic signatures in combination with multivariable logistic regression, automatic segmentation with AI methods were compared with manual segmentation through an integrated model. Dice similarity coefficients of automatic kidney segmentation were determined to be 0.894 and 0.881 for training and test sets respectively, illustrating the feasibility and potential to assess the renal function of each kidney using plain CT and AI methods.
Outcomes and Implications
Because separate renal function assessments are typically performed using a nuclear renal scan, the use of radioactive tracers to monitor filtration and excretion of each kidney is necessary. By using other methods such as contrast-enhanced CT in combination with AI models, both the radiation risk as well as cost is reduced. Consequently, this increases both procedural safety and efficiency, indicating CT and AI deep learning models as a potential non-invasive imaging technique.