LB1036 Artificial intelligence for the automated assessment of psoriasis severity

PASI score is globally used to assess disease activity of psoriasis. However, it relatively complicated and time-consuming, the will vary due inconsistent subjectivity between dermatologists. Therefore, an AI system capable assessing psoriasis severity be useful. We showed progress our research at SID meeting last year. Recently, we have established a novel platform for evaluation by using deep convolutional neural networks. 705 images trunk’s front back were in research. Considering small number images, data augmentation techniques expand data. A expert’s scores as teacher Various network models hyperparameters adjusted five-fold cross validation. From these adjustments, discovered that fine-tuning Imagenet2012-pretrained InceptionV3 whose linear layer was replaced two-layer perceptron (30 hidden units five output units) exhibited best performance. To validate learning system, 10 selected test sets excluded from training sets. The assessment almost consistent with clinical severity. examined whether assistance would affect human scoring. 13 dermatologists nine medical students invited evaluators. Mean absolute differences standard deviation among evaluators reduced assistance. Moreover, evaluator's got close teacher’s AI’s developed simply uploading single image. An easy-to-use scoring help patients