Research On Medical Image Segmentation Algorithm And Related Applications

Medical image segmentation is critical for disease diagnosis,surgical planning and treatment assessment.However,manual delineation is a time-consuming process and prone to human errors.Therefore,an accurate,rapid and reproducible automatic segmentation technique is urgently needed in clinical applications.With the advent of digital and intelligent age,automatic medical image segmentation technology has also achieved rapid development.In this paper we mainly studied two theories and applications related to automatic medical image segmentation.In this paper,we proposed a modified cascaded 3D U-Net method to segment brain tumor in multi-modality MRI image for imbalanced classes.The data used in experiments comes from BraTS 2019 training set.We use 80% of the dataset for training and the remaining 20% for validation.In pre-processing step,nonbrain regions are removed.We used a cascaded 3D U-Net to segment the brain tumors.The first 3D U-Net uses four modalities images as inputs,and outputs the mask of whole tumor(WT).The second 3D U-Net only uses T1 ce,T2 and Flair images and the patches which comprise all three tumor classes are kept for training to segment the WT into three substructures: edema(ED),tumor core(TC)and enhancing tumor(ET).P-ReLu and focal loss were used as the activation and loss function,providing the activation of negative features and the reduction of the relative loss for wellclassified examples.The batch size was 2 and the group normalization was used to stabilize the computed statistics.The initial segmentation of WT helps suppress false-positive classifications in non-tumorous areas.Combined with the use of focal loss,the proposed method mitigates the effect of unbalanced data.The mean dice scores on the validation dataset are 0.731,0.908,0.822 for ET,WT,and TC,respectively.The HD95 distance on the validation dataset are 4.986 mm,5.620 mm,6.339 mm for ET,WT,and TC,respectively.Besides,an automatic segmentation method was proposed to segment retinal layers mainly based on graph theory.In this method,the retinal segmentation task is transformed into graph partition problem.Firstly,the vertical gradient and the retinal layer thickness information in previous B-scans was used to construct the function for the weight between nodes in a graph.For more obvious boundaries such as ILM,BMEIS and OB-RPE,we only used gradient information to calculate the weights between nodes.For the fuzzy boundary,because the retinal structure changes little between two adjacent images and the lesion area does not cause the retinal layer thickness to change greatly,we added the retinal layer thickness information from previous B-scan to constrain the retinal layer boundaries.Then,flexible search regions are also constructed by the previous one B-scan information in which the Dijkstra's algorithm was used to search the shortest path.And the shortest path is the retinal layer boundary.The proposed method was evaluated on 5 CSC cases(640 Bscans)and 5 normal cases(640 B-scans).For the image with CSC,the overall mean absolute boundary positioning difference is 9.86±4.52?m and the mean absolute thickness difference is 10.92±3.86?m.