Design And Implementation Of Image Analysis And Processing System For Dermatoscope

Skin cancer is a major public health issue that human beings need to get over,as more than five million new cases are diagnosed globally every year.In 2015,the global incidence of melanoma exceeded 350,000 with nearly 60,000 deaths.Although the mortality rate was high,melanoma survival rate was higher than 95%at the time of early detection.Dermatologists usually screen and diagnose melanoma at an early stage by naked eyes and histopathological biopsy[1].However,even experienced experts will misdiagnose at a relatively high rate merely by naked eyes,resulting in a lot of unnecessary pathological biopsy.Although pathological biopsy is the diagnostic gold standard,it often causes economic stress and physical and mental harm to patients.The dermatoscope assisted diagnosis technology has been developing rapidly at home and abroad in recent years,brought great convenience and efficiency improvement to clinicians,and in the meantime,further improved the accuracy of clinical diagnosis combined with artificial intelligence following the vigorous development of artificial intelligence in China.There is a corresponding relationship between dermoscopy features of skin and histopathology,which bridging clinic and pathology[2].How to design a system to analyze and process the dermatoscope image simply,conveniently and quickly,improve doctors’ diagnostic ability and shorten their diagnostic time,is a problem worth solving and of great significance.This thesis design and implement a dermoscopy image analysis and processing system based on deep learning,digital image processing technology and Electron client frame technology.The main functional modules and innovations of the system are as follows:1.The system introduces the quality evaluation module,which realizes fuzzy quality evaluation,uneven light quality evaluation and hair occlusion quality evaluation.The blurred picture,uneven light and serious hair occlusion due to improper operation of users or exuberant hair of patients often makes the picture difficult to analyze and process.This function module will first inform the lower quality picture.This function improves the user experience,users do not need to measure the quality of a skin mirror image by the naked eye.At the same time,less low quality image improves the accuracy of the follow-up model,reduces the misdiagnosis rate,and improves the specificity of diagnosis.2.The system introduces deep learning algorithms to make analysis of dermatoscope pictures in focus segmentation,focus dermoscopy attribute segmentation,dermatology identification,and benign and malignant prediction of focus.The function of focus segmentation and the function of focus dermoscopy attribute segmentation were both obtained by training with U-net neural network.Dermatoscopic recognition and benign and malignant focus recognition were obtained by training with Google-Net neural network.And the system uses Jaccard index and accuracy index to evaluate the model.The boundary of lesions and the area of dermoscopic attributes can provide a reliable basis for the follow-up analysis of skin lesions;the recognition of skin diseases and the prediction of benign and malignant lesions can provide a diagnostic reference for doctors and patients,which can make patients be found in time and greatly reduce the mortality of patients.3.The system introduces the enhancement module of dermatoscope image,and obtains the segmentation results of focus and the contour of the segmentation of focus dermatoscope attributes through digital image processing technology,so as to make each area more clearly presented..It uses the top hat closure algorithm to extract the hair,and uses the image repair algorithm to remove the hair.This function optimizes the user’s experience of reading film,makes the user more clearly observe the focus area,and the hair removal function can also improve the accuracy of model recognition.4.The system introduces the operation prediction result module,and uses the cornerstonejs component to provide users with the functions of image scaling,moving,window width and position adjustment,magnifying glass,ellipse annotation,rectangle annotation,brush,eraser,rotation,color reversal and mirror image inversion.This thesis finally realizes the dermatoscope image analysis and processing system.Users can carry out automatical analysis and processing of the dermatoscope image after importing it,and all the analysis results are displayed in the form of pictures or charts.The graphic visual interface greatly improves user experience,and the rapid analysis and processing function which takes several seconds immensely facilitates doctors to read X-ray,the deep learning model with high Jaccard index and high accuarcy improves diagnostic specificity,and avoids unnecessary biopsy.