H&E Feature Map Generation with Octave Convolution
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing image processing methods for H&E stained tissue samples are inefficient in terms of computational resources and processing time, particularly when determining low-frequency features.
Innovation Solution
The use of octave convolution in convolutional neural networks to generate high-frequency and low-frequency feature maps, where low-frequency feature maps have a lower spatial resolution, reducing the number of required convolution operations and improving processing efficiency.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If conventional convolutional neural networks process H&E stained tissue samples to extract features, then feature extraction can be performed, but processing time and computational resource consumption increase significantly
Solution Approach 1:
The patent segments the feature extraction process into two distinct pathways: a first pathway for extracting high-frequency features using standard convolution operations, and a second pathway for extracting low-frequency features using octave convolution with reduced spatial resolution. This segmentation allows each pathway to be optimized independently, with the low-frequency pathway operating at lower computational cost while the high-frequency pathway maintains full resolution for detailed feature detection.
Solution Approach 2:
The patent applies different processing qualities to different spatial frequency components. High-frequency features (detailed structures) are processed with full spatial resolution using conventional convolutions, while low-frequency features (general patterns) are processed with reduced spatial resolution using octave convolution. This local quality differentiation ensures that computational resources are allocated efficiently based on the specific requirements of each feature type.
2Measurement precision
If conventional convolutional neural networks perform convolution operations to extract features, then feature information can be obtained, but computational resource consumption increases
Solution Approach 1:
The computational process is segmented into two independent feature extraction pathways. The first pathway handles high-frequency features with full computational precision, while the second pathway handles low-frequency features with reduced computational precision. This segmentation enables the system to achieve the necessary feature detection accuracy without uniformly applying high computational cost to all feature extraction operations.
Solution Approach 2:
The patent changes the spatial resolution parameter when extracting low-frequency features. By operating at a reduced spatial resolution for low-frequency feature extraction, the system maintains the ability to detect general patterns and structures while significantly reducing the number of computational operations required, thereby lowering energy consumption.
3Measurement precision
If high spatial resolution is used for all feature maps, then detailed feature information is captured, but processing efficiency decreases
Solution Approach 1:
The patent implements local quality by maintaining full spatial resolution only where necessary - specifically for high-frequency feature extraction. For low-frequency features, the system intentionally reduces spatial resolution to improve processing efficiency. This localized application of high quality ensures that detailed structural information is preserved when needed, while overall processing efficiency is maintained through selective reduction of computational complexity.
Data Source
Figure 1
Figure 2
Figure 3A
AI summary
An apparatus and methods for determining features in an image of a Hematoxylin and Eosin (H&E) stained tissue sample. An apparatus can be configured to apply a machine learning model to the H&E stained tissue sample image to determine features in the image. Applying the machine learning model can include performing convolution operations on the H&E stained tissue sample image to generate a plurality of initial feature maps, applying octave-convolution-first-layer operations on the initial feature maps to generate initial high-frequency feature maps and low-frequency feature maps, applying octave-convolution operations on the high-frequency initial feature maps to generate refined high-frequency feature maps, applying octave-convolution operations on the low-frequency initial feature maps to generate refined low-frequency feature maps, and applying octave-convolution last layer operations on the refined high-frequency feature maps and the refined low-frequency feature maps to produce combined feature maps.