Res-CRANN Chromosome Classification via Attention Mechanism
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Conventional automatic chromosome classification methods are inefficient and lack accuracy, relying on features like length and banding profiles, and deep learning models, such as recurrent neural nets, struggle to effectively classify chromosomes due to their sequential nature and the need for precise visual feature extraction from small chromosome images.
Innovation Solution
A processor-implemented method using a Recurrent Neural Network (RNN) augmented with an attention mechanism, which generates feature maps, extracts feature vectors, and concatenates them to create a feature sequence for chromosome classification, focusing on Regions-of-interest (ROIs) specific to class labels, thereby improving classification accuracy.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Device complexity
If conventional automatic chromosome classification methods use length and banding profile features, then the classification process is simplified, but the classification accuracy deteriorates
Solution Approach 1:
The patent transforms the chromosome image into a sequence of feature vectors by changing the representation parameters from simple length and banding profiles to detailed visual features extracted through convolutional layers, enabling the RNN to learn sequential patterns that improve classification accuracy
Solution Approach 2:
The patent introduces an intermediary RNN component that processes the sequence of feature vectors extracted from chromosome bands. This intermediary layer bridges the gap between simple feature extraction and final classification, allowing the system to capture sequential dependencies while maintaining manageable complexity
2Measurement precision
If deep learning models process small chromosome images, then visual feature extraction capability is improved, but the computational efficiency deteriorates
Solution Approach 1:
The patent segments the chromosome image into multiple horizontal slices, each processed independently to generate feature vectors. This segmentation allows the model to focus on local band patterns while reducing the overall computational burden compared to processing the entire image as a single unit
Solution Approach 2:
The patent employs a dynamic RNN architecture that adaptively processes the sequence of feature vectors, adjusting its internal state to capture sequential dependencies. This dynamic processing enables efficient learning from the segmented features without requiring excessive computational resources
3Measurement precision
If Recurrent Neural Network processes sequential chromosome bands, then the ability to learn band sequences is improved, but the processing time increases
Solution Approach 1:
The patent extracts only the essential sequential information from chromosome bands by converting each horizontal slice into a feature vector and feeding it to the RNN. This extraction approach captures the critical band sequence patterns while discarding redundant information, reducing processing time
Solution Approach 2:
The patent applies partial action by processing only the most informative horizontal slices of the chromosome image rather than every possible feature. This selective processing maintains the ability to learn band sequences while avoiding excessive computation on less relevant data
Data Source
Figure 1
Figure 2
Figure 3a~3b
AI summary
Method and system for automatic chromosome classification is disclosed. The system, alternatively referred as a Residual Convolutional Recurrent Attention Neural Network (Res-CRANN), utilizes property of band sequence of chromosome bands for chromosome classification. The Res-CRANN is end-to-end trainable system, in which a sequence of feature vectors are extracted from the feature maps produced by convolutional layers of a Residual neural networks (ResNet), wherein the feature vectors correspond to visual features representing chromosome bands in an chromosome image. The sequence feature vectors are fed into Recurrent Neural Networks (RNN) augmented with an attention mechanism. The RNN learns the sequence of feature vectors and the attention module concentrates on a plurality of Regions-of-interest (ROIs) of the sequence of feature vectors, wherein the ROIs are specific to a class label of chromosomes. The Res-CRANN provides higher classification accuracy as compared to the state-of the- art methods for chromosome classification.