Semantic Segmentation Fusion for Multi-Scale Pathology Images
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Conventional image recognition methods using convolutional neural networks struggle to balance accuracy with segmentation precision when identifying cancer tissue regions in gastric pathological images, as these regions vary significantly in size and texture, leading to low recognition accuracy and loss of segmentation information.
Innovation Solution
An image processing method utilizing a backbone network connected in parallel with a pooling module and a dilated convolution module, each with different dilation rates, to extract feature maps and fuse results for improved semantic segmentation.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If a method based on spatial pooling is used to recognize large-size and small-size cancer tissue areas, then the recognition accuracy is improved, but the segmentation accuracy is reduced due to great loss of pathological image information
Solution Approach 1:
The patent divides the image processing into two separate parallel streams: a spatial pooling stream for recognizing cancer tissue areas of different sizes, and a dilated convolution stream for preserving segmentation accuracy. This segmentation allows each stream to specialize in its strength without compromising the other.
Solution Approach 2:
The patent merges the outputs of the spatial pooling stream and the dilated convolution stream through a fusion module. The spatial pooling result provides recognition accuracy for cancer tissue areas, while the dilated convolution result preserves segmentation accuracy, and their combination achieves both goals simultaneously.
2Manufacturing precision
If a method based on dilated convolution at a single dilation rate is used, then the segmentation effect is improved for cancer tissue regions in a certain size range, but the method cannot meet the size range of actual cancer tissue regions
Solution Approach 1:
The patent applies dilated convolution with multiple different dilation rates (e.g., 1, 2, 4, 8) in parallel within the dilated convolution stream. This allows the system to adapt to cancer tissue regions of varying sizes by adjusting the receptive field through different dilation rates, thereby covering a wide size range while maintaining segmentation precision.
3Productivity
If a method based on the fully convolutional network is applied to large-size and small-size cancer tissue areas, then the processing capability is improved, but the recognition accuracy is reduced
Solution Approach 1:
The patent segments the fully convolutional network into two parallel streams: spatial pooling for recognizing cancer tissue areas of different sizes, and dilated convolution for maintaining segmentation precision. This segmentation allows the system to maintain high processing capability while improving recognition accuracy through specialized processing in each stream.
Data Source
Figure 1
Figure 2~3
Figure 4~5
AI summary
An image processing method comprises: acquiring a target image; calling an image recognition model comprising a backbone network, a pooling module and a dilated convolution module connected to the backbone network and parallel to each other, and a fusion module connected to the pooling module and the dilated convolution module; using the backbone network in the image recognition model to extract features from the target image, using the pooling module and the dilated convolution module to respectively process a feature map output by the backbone network, so as to obtain a first result output by the pooling module and a second result output by the dilated convolution module, using the fusion module to fuse the first result and the second result, and outputting a model recognition result of the target image; and determining a semantic segmentation label map of the target image on the basis of the model recognition result.