Early intervention for patients with autosomal dominant polycystic kidney disease (ADPKD) depends on an accurate prognosis of renal function decline. The biomarkers and factors used currently are height-adjusted total kidney volume (htTKV), estimated glomerular filtration rate (eGFR), and patient age.
Manual measurement of kidney volume is time-consuming and subject to observer variability. According to Anish Raj, MSc, and colleagues at Heidelberg University, Mannheim, Baden Württemberg, Germany, combining automatically generated features from kidney magnetic resonance imaging with conventional biomarkers can enhance prognostic improvement.
The researchers reported on two deep-learning algorithms developed at the university. First, an automated kidney volume segmentation model is used to accurately calculate htTKV. Second, the model utilizes segmented kidney volumes, predicted htTKV, age, and baseline eGFR to predict chronic kidney disease (CKD) stages ≥3A, ≥3B, and a 30% decline in eGFR at 8 years after the baseline visit.
The approach combines a convolutional neural network and a multilayer perceptron. In a study cohort of 135 patients, the area under the curve (AUC) scores were 0.96, 0.96, and 0.95 for CKD stages ≥3A, ≥3B, and a 30% decline in eGFR, respectively.
“Furthermore,” the researchers said, “the algorithm achieved a Pearson correlation coefficient of 0.81 between predicted and measured eGFR decline. We extended our approach to predict distinct CKD stages after 8 years with an AUC of 0.97. The proposed approach has the potential to enhance monitoring and facilitate prognosis in ADPKD patients, even in the early disease stages.”
Source: Journal of Medical Physics