Deep-learning classifier with ultrawide-field fundus ophthalmoscopy for detecting branch retinal vein occlusion

AIM:Toinvestigateandcomparetheefficacyoftwomachine-learningtechnologieswithdeep-learning(DL)andsupportvectormachine(SVM)forthedetectionofbranchretinalveinocclusion(BRVO)usingultrawide-fieldfundusimages.METHODS:Thisstudyincluded237imagesfrom236patientswithBRVOwithamean±standarddeviationofage66.3±10.6yand229imagesfrom176non-BRVOhealthysubjectswithameanageof64.9±9.4y.Trainingwasconductedusingadeepconvolutionalneuralnetworkusingultrawide-fieldfundusimagestoconstructtheDLmodel.Thesensitivity,specificity,positivepredictivevalue(PPV),negativepredictivevalue(NPV)andareaunderthecurve(AUC)werecalculatedtocomparethediagnosticabilitiesoftheDLandSVMmodels.RESULTS:FortheDLmodel,thesensitivity,specificity,PPV,NPVandAUCfordiagnosingBRVOwas94.0%(95%CI:93.8%-98.8%),97.0%(95%CI:89.7%-96.4%),96.5%(95%CI:94.3%-98.7%),93.2%(95%CI:90.5%-96.0%)and0.976(95%CI:0.960-0.993),respectively.Incontrast,fortheSVMmodel,thesevalueswere80.5%(95%CI:77.8%-87.9%),84.3%(95%CI:75.8%-86.1%),83.5%(95%CI:78.4%-88.6%),75.2%(95%CI:72.1%-78.3%)and0.857(95%CI:0.811-0.903),respectively.TheDLmodeloutperformedtheSVMmodelinalltheaforementionedparameters(P<0.001).CONCLUSION:TheseresultsindicatethatthecombinationoftheDLmodelandultrawide-fieldfundusophthalmoscopymaydistinguishbetweenhealthyandBRVOeyeswithahighlevelofaccuracy.TheproposedcombinationmaybeusedforautomaticallydiagnosingBRVOinpatientsresidinginremoteareaslackingaccesstoanophthalmicmedicalcenter.