The Research On The Identification Of Skin Melanoma Based On Microscopic Hyperspectral Imaging

Cutaneous Malignant Melanoma is a kind of highly malignant skin tumors.Although there is no cure at current stage,early diagnosis can be helpful for patient to get best timing of surgery,thus to help them to prolong their postoperative survival time.Therefore,research on the early diagnosis of Cutaneous Malignant Melanoma is of great significance.Since there is no quantitative analysis in the pathologic diagnosis of Cutaneous Malignant Melanoma,in this paper,hyperspectral imaging technology is applied in the recognition and quantitative analysis to provide a new method for early diagnosis.According to the variety of sample raw data,in this paper,target recognition and quantitative analysis is done for three typical Cutaneous Malignant Melanoma samples.The content of this paper is as follows:Firstly,the segmentation problem of skin granular layer is studied based on KMNF method.In this method,KMNF is used to extract the main features and reduce the noise of hyperspectral images.Then,morphological filtering is used to extract the main morphological character of granular layer.Level set is subsequently applied to obtain the morphological outlet of granular layer.The experiment results show that the accuracy of this method is over eighty percent in extracting granular layer.Therefore,it is able to obtain clear outlet for invasion depth calculation,using this method.Secondly,the segmentation problem of Cutaneous Malignant Melanoma is studied using LSSVM with characteristic spectrum supervision(CSS-LSSVM).In this method,LSSVM is used to establish the segmentation model of Cutaneous Malignant Melanoma.Based on the established LSSVM model,some typical characteristic spectrum of the target is chosen as reference to re-segment the model.The experiments results show that the accuracy of this method is over eighty five percent in segmenting the Cutaneous Malignant Melanoma.Furthermore,combining with the above segmentation results of skin granular layer,the Melanoma Depth of Invasion is calculated.Finally,the segmentation of tutor region is studied using PSO-ELM algorithm.In this method,ELM is applied to establish the segmentation model of tutor region.Besides,to improve the segmentation accuracy of the model,the PSO algorithm is used to find optimum solution for ELM parameter.The experiment results show that,comparing with traditional SVM,the proposed PSO-ELM method can obtain higher segmentation accuracy and speed.Therefore,by using this method in the auto recognition of tutor area,the diagnosis efficiency will be improved.These above results also show that hyperspectral imaging technology can be used in the segmentation of Cutaneous Malignant Melanoma,provide quantitative data for its pathological diagnosis,and it has important significance for the clinical treatment and prognosis of Cutaneous Malignant Melanoma.