Automated Mapping Studies Of Tumors And Organs With Radiation Therapy

Objective:Accurate segmentation of tumors and organs in radiation therapy images is essential for clinical diagnosis,and manual segmentation is a time-consuming and subjective task.In recent years,great progress has been made in radiotherapy imaging tumor and organ segmentation technology based on deep learning.However,the intra-class heterogeneity of medical imaging caused by small-size tumors is not easy to segment and unclear boundary segmentation is still a problem that needs to be overcome in the segmentation process.In this study,an algorithm model for automatic segmentation of medical images based on deep learning was designed to improve the automatic delineation effect.Methods:A two-scale two-channel sliding window attention structure was proposed,and a feature fusion module was designed to combine coarse-grained and fine-grained features in the image,so that the learning of small-scale tumor delineation was more detailed.The ideas of locally weighted Gaussian attention and residual structure are introduced and integrated into the encoder and decoder in the network,which strengthens the learning of edge areas in the image in the network training stage and improves the recognition ability of irregularly shaped areas.Dice Similarity Coefficient(DSC),Positive Predictive Value(PPV),Hausdorff Distance(HD95),etc.were used as objective evaluation criteria to judge the accuracy of delineation and edge delineation of small-scale tumors.Results:In this study,DSC was 0.96 and HD95 was 4.36 in LiTS liver and liver tumor delineation experiments,DSC was 0.95 and HD95 was 1.74 in TCGA cranial tumor delineation experiments,and the DSC values of liver,left kidney,right kidney and spleen were 0.92,0.84,0.83 and 0.80 in the CHAOS abdominal multi-organ segmentation experiment,and HD95 values were 4.13,2.84,3.16,respectively.4.69,compared with the existing automatic sketching classical algorithms U-net,DeepLab v3+and PSPnet,the accuracy of segmentation is significantly improved,and the lowest HD95 value is obtained at the edge sketching,and the best effect is achieved.Conclusion:This paper proposes an algorithm framework that can be used for segmentation and automatic delineation of organs and tumors in medical images,and has been verified on CT and MRI public dataset slide images,through the evaluation of the algorithm in this paper and compared with other methods of automatic depiction of medical images,our algorithm can achieve the best performance in the segmentation of boundaries and segmentation problems at details,and can simulate clinical operations to automatically delineate organs and tumors,saving manpower and time resources compared with manual sketching.Improve the efficiency of clinical application delineation area and simplify the workflow of organ and tumor segmentation in clinical treatment.