Image classification and auxiliary diagnosis system for hyperpigmented skin diseases based on deep learning

Background and aimMelasma (ML), naevus fusco-caeruleus zygomaticus (NZ), freckles (FC), cafe-au-lait spots (CS), nevus of ota (NO), lentigo simplex (LS), are common skin diseases causing hyperpigmentation. Deep learning algorithms learn the inherent laws representation levels sample data can analyze internal details image classify it objectively to be used for diagnosis. However, deep that assist clinicians in diagnosing hyperpigmentation conditions lacking.MethodsThe optimal deep-learning recognition algorithm was explored auxiliary diagnosis hyperpigmented disease. Pretrained models, such as VGG-19, GoogLeNet, InceptionV3, ResNet50V2, ResNet101V2, ResNet152V2, InceptionResNetV2, DesseNet201, MobileNet, NASNetMobile were images six diseases. The best developing an online clinical system selected by using accuracy area under curve (AUC) evaluation indicators.ResultsIn this research, parameters above-mentioned ten 18333510, 5979702, 21815078, 23577094, 42638854, 58343942, 54345958, 3235014, 4276058, respectively, their training time 380, 162, 199, 188, 315, 511, 471, 697, 101, 144 min respectively. respective accuracies set 85.94%, 99.72%, 99.61%, 99.52%, 98.84%, 99.13%, 99.61%. rates test 73.28%, 57.40%, 70.04%, 71.48%, 68.23%, 71.11%, 71.84%, 70.39%, 43.68%, Finally, areas AUC curves 0.93, 0.86, 0.91, 0.92, 0.82, respectively.ConclusionsThe experimental parameters, time, accuracy, above models suggest MobileNet provides a good application prospect skin.