Skin Disease Detection and Classification of Actinic Keratosis and Psoriasis Utilizing Deep Transfer Learning

Skin diseasescanarisefrominfections,allergies,geneticfactors,autoimmunedisorders,hormonal imbalances, orenvironmentaltriggerssuchassundamageandpollution.SkindiseasessuchasActinic KeratosisandPsoriasiscanbefatal.Thesearetreatableifidentifiedearly.However,itsdiagnosticmethods are expensiveandnotwidelyaccessible.Inthisstudy,anovelandefficientmethodfordiagnosingskin diseases usingdeeplearningtechniqueshasbeenproposed.Thisapproachemploysmultiplemodified ConvolutionalNeuralNetwork(CNN)modelslikeDenseNet169,DenseNet201andVGG16.Softvoting ensemble strategyisappliedtocombinethestrengthsofindividualmodelstogetbetterresult.These models includeseveralconvolutionallayers.ThemodelshavebeenemployedusingImageNetweightsand modified toplayers.Thetoplayersaremodifiedbyfullyconnectedlayersandafinalsoftmaxactivation layertoobtaintheresult.Thedatasetanalyzedispubliclyavailableandtitled“SkinDiseaseDataset”.The CNN architecturedoesnotincludeaugmentationbydefault;dataaugmentationistypicallyperformed duringpreprocessingpriortomodeltraining.Theproposedmethodologyachieved93.11%accuracyusing the ensemblestrategy,demonstratingreliabilityinclassifyingskindiseases.Themodifiedpre-trained models showedpromisingresults,increasingitspotentialforreal-worldapplications.