Intelligent Analysis System Of The Melanonychia Image Based On Deep Learning Image Segmentation

Dark pigment(pigmented nevus)on nails is a very common phenomenon,which is called melanonychia and belongs to the skin diseases.The occurrence of melanonychia could be caused by malignant melanoma.Malignant melanoma is one kind of cancer with high mortality rate,so its early diagnosis and treatment is very important.The diagnosis of melanonychia mainly refers to the ABCDE rule,which contains several indicators of the pigmented nevus,including asymmetry,border,color,diameter and so on.The current research of the melanonychia based on on artificial intelligence faces the following problems:(1)There is no high-quality public dataset,which leads to the insufficient data support;(2)Lack of relevant research,most of the research is based on end-to-end classification algorithms,which is not interpretable;(3)The melanonychia image includes dermoscopy image and mobile phone image,which has the problem of distribution bias;(4)Lack of landing applications.Based on the above problems,this paper mainly carry out three following tasks:(1)This paper carried out the research on the dermoscopy image analysis algorithm of melanonychia,aiming at the problem of lack of dataset,established a dermoscopy image dataset of the melanonychia,and proposed a dermoscopy image analysis system based on image segmentation algorithm,which is composed of an image segmentation Sub-algorithm and an indicator analysis Sub-algorithm.The fuction of the image segmentation Sub-algorithm is extracting the nail area and discoloration area of the melanonychia dermoscopy image.To achieve that,this paper proposed an improved image segmentation algorithm based on Attention UNet.By adding a feature riching module and a boundary enhancement module,the feature utilization capability and the edge segmentation effect of the network is improved.The experimental result shows that the proposed algorithm can effectively segment the melanonychia dermoscopy image,and the segmentation effect is better than the Attention UNet.Following the image segmentation Sub-algorithm,the indicator analysis Sub-algorithm quantitatively analyzes the specific indicators according to the ABCDE rule.Through the doctors’ evaluation,the analysis system has good interpretability.(2)This paper carried out the research on the phone image analysis algorithm of melanonychia,aiming at the problem of lack of dataset,established a melanonychia mobile phone image dataset.Aiming at the difference in the visual representation distribution between the image taken by the mobile phone and the dermoscopy,a segmentation algorithm for the mobile phone image of the melanonychia based on domain adaptation was proposed.First,perform data preprocessing,using the automatic positioning module to find the block diagram of the nail position and cut it,then take the nail-black line dermoscopy image as the source domain(with label)and the melanonychia mobile phone image as the target domain(without label),then use the adversarial-based domain adaptive image segmentation algorithm to reduce the Using domain adaptive image segmentation algorithm to reduce the difference between domains in the output space,achieving the unsupervised segmentation of mobile phone images.The experimental results show that the algorithm can effectively segment mobile phone images without labeling,and it performs better than the comparison network.(3)This paper designed an intelligent analysis application system for melanonychia mobile phone image.The application system consists of an Android-based mobile App and a server.People can obtain the nail image by using the App and upload it to the server.The server obtains the image and calls the melanonychia image analysis algorithm,then returns the analysis result of the algorithm to the mobile phone.This application system helps patients to conduct self-condition analysis in the early stage and save medical resources.