Early Detection of Ureteropelvic Junction Obstruction Using Signal Analysis and Machine Learning: A Dynamic Solution to a Dynamic Problem
Abstract
Purpose:
We sought to define features that describe the dynamic information in diuresis renograms for the early detection of clinically significant hydronephrosis caused by ureteropelvic junction obstruction.
Materials and Methods:
We studied the diuresis renogram of 55 patients with a mean ± SD age of 75 ± 66 days who had congenital hydronephrosis at initial presentation. Five patients had bilaterally affected kidneys for a total of 60 diuresis renograms. Surgery was performed on 35 kidneys. We extracted 45 features based on curve shape and wavelet analysis from the drainage curves recorded after furosemide administration. The optimal features were selected as the combination that maximized the ROC AUC obtained from a linear support vector machine classifier trained to classify patients as with or without obstruction. Using these optimal features we performed leave 1 out cross validation to estimate the accuracy, sensitivity and specificity of our framework. Results were compared to those obtained using post-diuresis drainage half-time and the percent of clearance after 30 minutes.
Results:
Our framework had 93% accuracy, including 91% sensitivity and 96% specificity, to predict surgical cases. This was a significant improvement over the same accuracy of 82%, including 71% sensitivity and 96% specificity obtained from half-time and 30-minute clearance using the optimal thresholds of 24.57 minutes and 55.77%, respectively.
Conclusions:
Our machine learning framework significantly improved the diagnostic accuracy of clinically significant hydronephrosis compared to half-time and 30-minute clearance. This aids in the clinical decision making process by offering a tool for earlier detection of severe cases and it has the potential to reduce the number of diuresis renograms required for diagnosis.
References
- 1 : Hydronephrosis: prenatal and postnatal evaluation and management. Clin Perinatol2014; 41: 661. Google Scholar
- 2 : The Society for Fetal Urology consensus statement on the evaluation and management of antenatal hydronephrosis. J Pediatr Urol2010; 6: 212. Google Scholar
- 3 : The long-term followup of newborns with severe unilateral hydronephrosis initially treated nonoperatively. J Urol2000; 164: 1101. Link, Google Scholar
- 4 : Inter-observer reproducibility in reporting on renal drainage in children with hydronephrosis: a large collaborative study. Eur J Nucl Med Mol Imaging2008; 35: 644. Google Scholar
- 5 : Impaired drainage on diuretic renography using half-time or pelvic excretion efficiency is not a sign of obstruction in children with a prenatal diagnosis of unilateral renal pelvic dilatation. J Urol2003; 169: 1828. Link, Google Scholar
- 6 : Gravity-assisted drainage imaging in the assessment of pediatric hydronephrosis. Can Urol Assoc J2016; 10: 96. Google Scholar
- 7 : Diuretic renography with the addition of quantitative gravity-assisted drainage in infants and children. J Nucl Med2000; 41: 1030. Google Scholar
- 8 : Quantitative ultrasound for measuring obstructive severity in children with hydronephrosis. J Urol2016; 195: 1093. Link, Google Scholar
- 9 : Quantification of kidneys from 3D ultrasound in pediatric hydronephrosis. Conf Proc IEEE Eng Med Biol Soc2015; 2015: 157. Google Scholar
- 10 : Renal segmentation from 3D ultrasound via fuzzy appearance models and patient-specific alpha shapes. IEEE Trans Med Imaging2016; 35: 2393. Google Scholar
- 11 : Prediction of clinical outcomes in prenatal hydronephrosis: importance of gravity assisted drainage. J Urol2017; 197: 838. Link, Google Scholar
- 12 : B-spline signal processing. I. Theory. IEEE Trans Signal Process1993; 41: 821. Google Scholar
- 13 : Probability and random processes. J Am Stat Assoc1985; 80: 788. Google Scholar
- 14 : Support-vector networks. Mach Learn1995; 20: 273. Google Scholar
- 15 : Down syndrome in diverse populations. Am J Med Genet A2017; 173: 42. Google Scholar
- 16 : Computer-aided renal cancer quantification and classification from contrast-enhanced CT via histograms of curvature-related features. Conf Proc IEEE Eng Med Biol Soc2009; 2009: 6679. Google Scholar
- 17 : Digital facial dysmorphology for genetic screening: hierarchical constrained local model using ICA. Med Image Anal2014; 18: 699. Google Scholar
- 18 : Transitional neonatal hydronephrosis: fact or fantasy?. J Urol1986; 136: 339. Link, Google Scholar
- 19 : The “well tempered” diuretic renogram: a standard method to examine the asymptomatic neonate with hydronephrosis or hydroureteronephrosis. A report from combined meetings of The Society for Fetal Urology and members of The Pediatric Nuclear Medicine Council—the Society of Nuclear Medicine. J Nucl Med1992; 33: 2047. Medline, Google Scholar
- 20 : Antenatal detection of pelviureteric junction stenosis: main controversies. Semin Nucl Med2011; 41: 11. Google Scholar
- 21 : Key variables for interpreting 99mTc-mercaptoacetyltriglycine diuretic scans: development and validation of a predictive model. AJR Am J Roentgenol2011; 197: 325. Google Scholar
- 22 : Value of supranormal function and renogram patterns on 99mTc-mercaptoacetyltriglycine scintigraphy in relation to the extent of hydronephrosis for predicting ureteropelvic junction obstruction in the newborn. J Nucl Med2003; 44: 725. Google Scholar
