INTERPRETATION OF CAPTOPRIL TRANSPLANT RENOGRAPHY USING A FEED FORWARD NEURAL-NETWORK

Severe renal artery stenosis (RAS) is a relatively uncommon complicati on after renal transplantation but is a curable cause of hypertension, which demands reliable early diagnosis to reduce morbidity, mortality and graft loss. Captopril renography has been used for a number of ye ars as a method of detecting RAS but controversy still exists as to th e diagnostic accuracy of this test and as to the most appropriate inte rpretation criteria with which to establish a positive result. Methods : This report presents the results of using artificial neural networks to impartially assess these interpretation criteria. Data comprised 3 1 Tc-99m-MAG3 captopril renography investigations undertaken on hypert ensive renal transplant patients with a suspected diagnosis of RAS. Ea ch renogram study was correlated with an arteriogram as the ''gold sta ndard''. Training of the network was performed using the round-robin t echnique. Results: An accuracy of 95% could be achieved by considering perfusion index, time-to-peak activity, accumulation index and excret ion index for both pre- and post-challenge studies. This varied as the parameters were either included or excluded. Conclusion: Artifical ne ural network analysis is a useful technique to evaluate the most appro priate criteria for interpreting captopril transplant renography inves tigations.