Artificial Intelligence-assisted Diagnosis Of TCM Syndromes Of Psoriasis Based On Dermoscopic Feature

Background:Traditional Chinese medicine(TCM)has a good effect on psoriasis,but the discrimination of syndrome still depends on the doctor’s subjective judgment,which is greatly affected by the environment,light,personal factors of the doctor.As a non-invasive transparent electron microscope of the skin surface,dermoscopy can capture information of different levels of skin lesions more comprehensively.It has been confirmed that dermoscopic images can contribute to TCM syndrome classification of psoriasis vulgaris.While artificial intelligence(AI)has developed rapidly in the field of medical image recognition,there has also been a certain research foundation in the classification recognition of dermoscopic images.Objective:To construct an artificial intelligence image recognition program to classify and recognize dermoscopic images of three basic TCM syndromes of psoriasis vulgaris(blood-heat syndrome,blood-stasis syndrome and blood-dryness syndrome),and to establish an image classification system that can be applied in clinical practice.Methods:Collect psoriasis vulgaris patients with blood-heat syndrome,blood-dryness syndrome and blood-stasis syndrome diagnosed by dermatology department of Beijing Traditional Chinese Medicine Hospital.Dermatoscopic images of typical skin lesions of patients with each syndrome were collected by using a handheld digital dermatoscope.Classify and process the images,and build image recognition program which refined by improving the criteria of image capture,segmentation methods and updating the learning framework.Finally,an intelligent recognition system for clinical use will be identified.Results:955 pictures were collected in the first phase of the experiment.VGG19,Efficientnet-b3,Xception,Inceptionv4,Nasnetalarge,Nasnetamobile and other in-depth learning frameworks were selected as training set to train model performance.Using 105 pictures as test set.the recognition accuracy of each model is at the 44-62%level,among which the data performance of Efficientnet-b3 model is better than that of other models(61.91%).After review,the second phase of the experiment restricted the method of dermatoscopy.A total of 803 pictures were collected.After image segmentation and normalization calculation,the recognition accuracy of Efficientnet-b3 increased from 61.905%to 93.91%.In this way.the framework of Efficientnet-b6 and Efficientnet-b7 was used for training.The accuracy of model recognition was increased to 97.4%,and the final recognition program was formed with Efficientnet-b7.Conclusion:Dermoscopy can observe the skin lesion features of epidermis and superficial dermis.which is helpful for TCM syndrome differentiation of psoriasis vulgaris.An artificial intelligence image recognition model based on Efficientnet is constructed.The model is used to classify the dermoscopic images of psoriasis vulgaris according to TCM syndrome differentiation.The results are in good agreement with the judgement of clinicians.It provides theoretical and technical support for standardized diagnosis and telemedicine of psoriasis vulgaris.