Semantic Segmentation Method Of Multiple Myeloma Cell Image Based On TMMC-DeeplabV3+

Multiple myeloma is one of the three major diseases of the blood system which seriously threatens patient’s life and health.Segmentation of multiple myeloma cell images is a key step for doctors to diagnose multiple myeloma.Traditional detection of multiple myeloma cell images relies on naked eye observation.The fatigue degree and professional degree of pathology experts will affect the accuracy of detection results.Pathology experts can’t provide high-quality test results quickly and stably.With the development of computer image technology,cell image segmentation method has been used in the diagnosis of multiple myeloma.Cell image segmentation method has gradually replaced the heavy work of artificial naked eye detection.In current existing multiple myeloma cell image semantic segmentation model,the time of multiple myeloma cell image segmentation is long,and the low accuracy cause by oversegmentation and under-segmentation in multiple myeloma cell image segmentation.Aiming at the above problems,this study proposes a semantic segmentation model of multiple myeloma cell image based on TMMC-Deeplab V3+algorithm.The main research work as follows:Firstly,a multiple myeloma cell image semantic segmentation algorithm based on Multi-level CBAM Deeplab V3+(MC-Deeplab V3+)has been proposed.Based on the indepth analysis of the low segmentation accuracy of the current existing multiple myeloma semantic segmentation model,a multiple myeloma cell image semantic segmentation algorithm based on MC-Deeplab V3+ has been proposed.In this algorithm,the convolution block attention block(CBAM)has been used to optimize the weight of Deeplab V3+ algorithm.In this algorithm,the atrous convolution of Atrous Spatial Pyramid Pooling(ASPP)has been reconstructed by multi-level cascade,the pooling layer of ASPP has been replaced by the Strip Pooling(SP)layer to expand the receptive field of ASPP.Secondly,a multiple myeloma cell image semantic segmentation algorithm based on Transfer Mobile Net V3 Multi-level CBAM Deeplab V3+(TMMC-Deeplab V3+)has been proposed.Aiming at the problem for the semantic segmentation model of multiple myeloma cell image based on MC-Deeplab V3+algorithm has poor real-time performance,a multiple myeloma cell image semantic segmentation algorithm based on the TMMCDeeplab V3+ has been proposed.In this algorithm,the Mobile Net V3 network has been used to replace the Xception network as the backbone extraction network of MCDeeplab V3+ algorithm to reduce the amount parameter of semantic segmentation model.Thirdly,the experimental verification of the multiple myeloma cell image semantic segmentation algorithms has been carried out.The experimental data-set has been processed successively through image selection,image labeling,label-file format conversion,data enhancement and data set delimitation.Based on the training-set and verification-set,a multiple myeloma cell image semantic segmentation model based on TMMC-Deeplab V3+algorithm has been constructed,and the iterative process of the semantic segmentation model has been compared.Based on the test-set,the semantic segmentation model ablation experiment,the semantic segmentation model contrast experiment,and the semantic segmentation result visualization contrast experiment has been carried out.Finally,a multiple myeloma cell images semantic segmentation interactive operation platform has been built.On the basis of the semantic segmentation model of multiple myeloma cell image,the interactive operation platform of multiple myeloma image semantic segmentation has been developed by Python language and the Thinter development platform development tool.The operation interactive platform page design has been succinct and handy to operate.The operation interaction platform can perform pixel-level semantic segmentation of multiple myeloma cell images.The operation interface of interactive operation platform login and multiple myeloma cell image semantic segmentation has been displayed.Figure [53] Table [5] Reference [87]...