Design And Implementation Of Intelligent Auxiliary Diagnosis System For Melanoma Based On Deep Learning

Skin melanoma tumors are currently the most common severe disease in the world,and they are a disease that can be cured at an early stage.According to incomplete statistics,thousands of people unfortunately fall ill each year.At present,the main diagnosis method of melanoma is based on dermatoscope artificial visual observation,but affected by the doctor’s medical skills and experience,the diagnosis accuracy rate is about 75% to 80%,and the diagnosis efficiency is low.Therefore,it is of practical significance to carry out early detection and accurate diagnosis of melanoma.The thesis comparatively studied the recognition and diagnosis methods for melanoma dermoscopic images at home and abroad,and proposed a multi-modal neural network algorithm that fused metadata and image data.Metadata is the feature vector that extracts the basic information of the patient,the location of the lesion collection,the resolution and quantity of the image through the perceptron learning model;the image data is the feature vector extracted through the CNN model,and the two feature vectors are fused and mapped to obtain the disease classification As a result,it is used for early diagnosis of melanoma tumors.The paper collects and sorts out a total of 58457 sample data of ISIC 2019 and ISIC2020 mixed data sets.The training samples and test samples are divided according to a 4:1 ratio,and the multi-modal algorithm and convolutional neural network method proposed in this article are used for comparative experimental research.,The results show that the melanoma tumor-assisted diagnosis classification model constructed using the algorithm of this paper can increase the AUC value by about 1%,which has a certain value in use.On this basis,the paper designs an adaptive multi-model fusion algorithm,which can set the number of fusion models and obtain the best classification result fusion model through exhaustive method.The ISIC 2019 and ISIC2020 mixed data sets were used for comparative experiments to analyze the influence of the number of fused multiple models on the performance of auxiliary diagnosis and classification of melanoma tumors.Experimental results show that the use of the best model fusion of 11 models can increase the classification accuracy of melanoma tumors by about 0.7% compared with the best single model,but the reasoning time will increase by 9 times.Flexible use of the model obtained by the adaptive model fusion algorithm can improve the efficiency of early diagnosis of melanoma tumors.Based on the research of the above two algorithms,a melanoma tumor intelligent auxiliary diagnosis system is designed and implemented,which has five functions: user registration and login,dermatoscope image data upload,deep learning training,classification result reasoning,and diagnosis result display.The test results show that the system has a certain use value.