Diagnostic Models And Technology Research For Melanoma Detection Based On Seven-point Checklist

Melanoma is caused by malignant abnormal proliferation of human tissue melanin cells，Often occurred in the the skin surface.In the death cases of skin cancer,about2/3of deaths arecausesd by malignant melanoma.The cancer often has spread to some tissues and organs,evenin patients with systemic,because early diagnosis of malignant melanoma is difficult，with highdegree of malignancy，easy spread and poor prognosis. Early diagnosis and early treatment isthe best way to cure the disease.Therefore, Early diagnosis is important for improving the curerate for melanoma.Seven point checklist accurately describes local texture features，structural features andcolor features in the diagnosis principles of melanoma, had been more widely used.Since itmainly describes microscopic features of melanoma, and is difficult to establish mathematicalmodel and to quantitative, so now assisted diagnosis based on7piont checklist is less. In orderto improve the accuracy and efficiency of the assisted diagnosis, providing more reliableinformation to the dermatologist, carrying out aided diagnosis study based on7point checklistis imperative.Irregular streaks, blue white veil, and regression structure are three important sub-featuresof7point checklist in which irregular streaks are micro-structural features, regression structuresand blue white veil belong color feature. About mathematical description of three sub-modelsand automatic detection methods, the paper conducted intensive study.The main contributions of of this paper are as follows:1) In the depth study on the diagnosis of melanoma based on7-point checklist，againstirregular streaks on the surface of melanoma，this paper proposed an extraction method basedon sub-block tumor direction histogram analysis and two-dimensional surface of Gabor filterbank, and gives macroscopic and microscopic measurement methods.The experimental resultsshow that the proposed method can well describe the streaks features on the surface ofmelanoma,and significantly improved the accuracy of streaks detection。.2) In the study on the the regression structure, this paper constructs a two-dimensionalcolor model based on G, B color components, with a statistical method on the representativesample of pixels. To obtain the regression features of melanoma accurately, and gives anaccurate measurement method.Experiments results show that the model could obtain regressionstructure with a high specificity. 3) In the study of blue white veil, the color of blue white veil changed complexly,so thatsimple color model is difficult to effectively obtain the feature region.Based on these analysis，in the paper a picture of a skin tumor with RGB model is changed to a lower dimensionalmodel using PCA analysis，simplifing the data and irrelevant，and conduct a model withprincipal component. The method is shown to produce effective results.4) Based on the principles of7-point checklist and scoring standards, the paper designscomprehensive evaluation system on sub-score. Firstly, get the irregularities streaks, blue whiteveil and regression structures, and extract their respective characteristic values as the input ofSVM classifier. Then the existence of the three features based on sub-SVM classification resultsis judged and diagnostic score of each sub-features feature is given. Lastly, summation of thetotal score determines the state of the melanoma.The system is fully intergrated with multiplefeatures, and the accuracy of aided diagnosis of melanoma is effectively improved.