Medical image processing method, device, electronic device and storage medium
A medical image and processing method technology, applied in the field of image processing, can solve the problems of difficult to accurately identify the boundary of the lesion area, low segmentation accuracy and low contrast of the ischemic lesion area, and achieve effective segmentation, good reinforcement and restraint. , the effect of improving the accuracy
- Summary
- Abstract
- Description
- Claims
- Application Information
AI Technical Summary
Problems solved by technology
Method used
Image
Examples
experiment example 1
[0143] Experimental Example 1 Performance Evaluation and Comparative Experiment
[0144] We have evaluated the technical scheme in the embodiment of the present invention (such as Figure 5 shown) and the performance of the comparison models, where the comparison models include U-Net, U-Net++, PSP-Net, DeepLabv3+, SF-Net, Inf-Net, CE-Net, OC-Net and the newly proposed Swin-UperNet ( transformer model). In addition, unless otherwise specified, the technical solutions in the embodiments of the present invention use ResNet-16 as the backbone network for feature extraction. The segmentation performance results of the inventive and comparative models are shown in Table 1.
[0145]As shown in Table 1, the technical solution in the embodiment of the present invention has reached the highest value in terms of Dice index, IoU and sensitivity, which are 1.3%, 1.2% and 3.1% higher than the second-ranked model respectively. In terms of specificity, there was little difference between a...
Embodiment 2
[0149] Embodiment 2 Backbone network replacement comparison experiment
[0150] This experimental example is in Figure 5 On the basis of the technical solution, the feature extraction module was replaced for experiments, and then the replaced technical solution was compared with other models with the same backbone network. The results are shown in Table 2. Among them, IS-Net basically adopts Figure 5 In the technical scheme, only different backbone networks are used as its feature extraction modules.
[0151] Table 2
[0152]
[0153] As shown in Table 2: For each backbone network, the IS-Net of the present invention is compared with a benchmark test mode, that is, IS-Net with backbone network ResNet-16 compared with DeepLabv3+, IS-Net with backbone network Res2Net Net and Inf-Net comparison, IS-Net with backbone network Swin-T and Swin-UperNet comparison. Figures in parentheses illustrate the improvement of IS-Net over comparable models with the same backbone network...
experiment example 3
[0154] Experimental example 3 fusion strategy replacement comparison experiment
[0155] This experimental example is in Figure 5 Based on the technical solution, different fusion strategies are used to replace the feature pyramid module. The schemes with different fusion strategies were used to segment the lesion area, and the segmentation performance was evaluated and compared. The results are shown in Table 3.
[0156] table 3
[0157] fusion strategy Dice (%) IoU(%) Sens.(%) edge constraints 67.5 56.7 74.4 marginal attention 66.5 55.3 71.8 FAM 67.3 56.1 74.3 FPN 67.0 55.8 73.6
[0158] Different integration strategies such as Figure 10 shown. Among them, FPN enhances the low-level high-level feature map by additively fusing the up-sampled high-level high-level feature map. FAM is a feature alignment module that enhances information dissemination between high-level high-level feature maps and low-level high-level fe...
PUM
Login to View More Abstract
Description
Claims
Application Information
Login to View More