Multi-Resolution CNN Fusion for Real-Time Tissue Lesion Detection
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing CAD systems struggle to process high-resolution medical images in real-time, integrate information from different scales, and handle varying object sizes in tissue lesions under a microscope, leading to inefficiencies and performance degradation.
Innovation Solution
A modified CNN system that processes images hierarchically, fusing information from different scales and resolutions using a cascade of CNNs and multi-layer perceptrons, allowing real-time processing and reduced computational cost.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If high-resolution microscopy images are processed to detect tissue lesions, then detection accuracy is improved, but processing time increases and real-time capability is lost
Solution Approach 1:
The patent divides the high-resolution microscopy image into multiple patches at different resolutions. A cascade of CNN processors handles these patches hierarchically, where lower-resolution processors handle coarse regions and higher-resolution processors handle detailed regions. This segmentation enables real-time processing by avoiding the need to process the entire high-resolution image at full detail simultaneously.
Solution Approach 2:
The patent introduces a multi-resolution dimension to image processing by creating a pyramid of images at different scales. This allows the system to process images at varying levels of detail, transforming the single-resolution processing bottleneck into a multi-level hierarchical structure that maintains real-time capability while preserving detection accuracy.
2Measurement precision
If separate CNN models are trained for each resolution scale, then performance at each resolution is optimized, but model size and computational cost increase significantly
Solution Approach 1:
The patent trains a single CNN model to handle multiple resolution scales universally. The model is designed with a cascade structure where the same network architecture processes images at different resolutions through hierarchical stages. This universal approach eliminates the need for separate specialized models for each resolution, reducing overall model size and computational complexity while maintaining performance across scales.
Solution Approach 2:
The patent merges multiple resolution processing paths into a single integrated CNN architecture. Instead of maintaining separate models for different resolutions, the system combines them into one hierarchical cascade where feature maps from lower resolutions feed into higher-resolution processing stages, consolidating computational resources and reducing total model complexity.
3Ease of manufacture
If local features are emphasized in CNN processing, then feature extraction is simplified, but global context information is lost leading to misinterpretation
Solution Approach 1:
The patent implements a nested hierarchical structure where low-resolution feature maps are processed first and then nested within higher-resolution feature maps. Each resolution level processes its local features while preserving the global context from lower levels. This nesting allows the system to maintain both local feature extraction simplicity and global context awareness, as higher-resolution features are built upon the contextual foundation of lower-resolution features.
Data Source
Figure 1
Figure 2
Figure 3
AI summary
Embodiments of the present disclosure include a method, device and computer readable medium involving receiving image data to detect tissue lesions, passing the image data through at least one first convoluted neural network, segmenting the image data, fusing the segmented image data, and detecting tissue lesions.