Study Of Automatic Detection Method Of Cancer Cell Based On Portable Microscope

Breast cancer,a kind of common female cancer,is deemed one of main diseases resulting in female death.Currently,most pathologists observe the status of cancer cell in the pathological section image via microscope in the visual observation method to diagnose breast cancer disease and achieve the pathological grading.However,such method’s accuracy rate in detecting and recognizing cancer cell is affected by the mental state of pathologists who bear high intensity,therefore,such method is greatly subjective.A kind of computer aided diagnosis method is proposed in this paper to assist pathologists in detecting and recognizing the breast cancer cell in the breast cancer pathology image with two kinds of dyeing manners.First,a portable cancer cell microscopic imaging system is designed to collect the pathologic cell images of breast cancer to address the large volume of microscopic imaging system and expensive cost.The test board with the USAF-1951 resolution ratio is adopted to test the system imaging performance,and the test results indicate that the resolution ratio of imaging system designed in this paper reaches 2.19m;Then the form and size of pathologic cell are compared,and the comparison results indicate that the imaging system designed in this paper can achieve the image acquisition of breast cancer cell.The design system in this paper has the following strengths: the volume structure of the imaging system is far less than that of the existing microscopic imaging system,and the volume of the imaging system is only203cm3;Compared with the high-end microscopic imaging system cost of over RMB ten thousand,the cost of design system in this paper declines greatly.Second,since the characteristics of breast cancer cell image concerning dyeing extraction in immunochemistry method(IHC)in the single color space aren’t enough to accurately finish the detection segmentation task of breast cancer cell,the characteristic equation is constructed based on the integration of multi-channel chromaticity characteristics to achieve the detection segmentation of breast cancer cell.In that method,the breast cancer cell image of IHC dyeing is pretreated and different mammary gland subsample cell images are cut;Then the chromaticity distribution information of subsamples in different channels is counted and analyzed,and the chromaticity characteristic equation is constructed as per the analysis results to achieve the detection segmentation of breast cancer cell;Later,the method based on connected domain is adopted to count the detected and recognized breast cancer cells.The experimental results indicate that the accuracy of the improved algorithm increases by 3.6% than that of unimproved algorithm.Eventually,due to small performance difference between mitotic cell and non-mitotic cell,it is very difficult for the traditional image segmentation method to extract enough characteristics.Therefore,the detection and segmentation method of mitotic cell image of breast cancer is researched,later,the Unet is selected as framework to raise the network MCSA-Unet composed based on attention mechanism and residual structure.Such network structure addresses the poor segmentation accuracy of mitotic cell image of breast cancer resulting from the insufficient extracted features of Unet network in the training process.Since the existing public mitotic cell dataset is very few,with weak annotation,and the dataset isn’t suitable for the network model in this paper although authoritative,the microscopic imaging system will be adopted to collect the mitotic cell image of breast cancer,and the pathology experts finish annotating mitotic cell;Later,the annotated mitotic cell image of breast cancer is made into dataset to provide data support for the later algorithm verification.The network algorithm is proposed in this paper as the semantic segmentation model,so,the Dice,Mean Pixel Accuracy(MPA),etc.are adopted as the evaluation indicator of MCSA-Unet network.The experiment results show that the indicators of algorithm MPA and Dice are 0.74 and 0.84.