Chromosome Segmentation via Crowdsourcing and Deep Learning
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Manual segmentation and classification of chromosomes in cytogenetics is laborious and time-consuming, especially for overlapping chromosomes, and existing automation techniques have limited success, requiring human intervention due to challenges like unsplit clusters and chromosome fragmentation.
Innovation Solution
A system combining crowdsourcing for initial chromosome segmentation and deep learning for classification, which includes preprocessing steps like straightening bent chromosomes and normalizing lengths, to improve accuracy and reduce dependency on experts.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If manual segmentation and classification of chromosomes is performed, then classification accuracy can be maintained, but time consumption and labor intensity increase significantly
Solution Approach 1:
The patent divides the chromosome analysis task into two segments: initial automated segmentation using image processing techniques, and subsequent classification using deep learning models. This segmentation allows the system to leverage automated methods for time-consuming tasks while maintaining accuracy through specialized neural networks trained on chromosome features.
Solution Approach 2:
The patent introduces a crowdsourcing platform as an intermediary to provide annotated chromosome images for training deep learning models. This intermediary layer enables the system to learn from diverse expert annotations, improving classification accuracy while reducing the need for manual intervention in routine analysis tasks.
2Productivity
If automated image processing techniques are used for chromosome segmentation, then time consumption is reduced, but segmentation accuracy deteriorates for overlapping chromosomes
Solution Approach 1:
The patent employs dynamic adjustment of processing strategies based on chromosome complexity. For simple non-overlapping chromosomes, fast automated image processing is used. For overlapping or complex cases, the system dynamically switches to more sophisticated segmentation approaches or flags them for expert review, thus maintaining both speed and accuracy.
Solution Approach 2:
The patent applies preliminary preprocessing steps such as image enhancement, contrast adjustment, and initial chromosome detection before main segmentation. This preliminary action prepares the data for more accurate segmentation by reducing noise and highlighting key features, thereby improving the performance of subsequent automated segmentation algorithms.
3Device complexity
If deep learning models are trained without preprocessing, then processing steps are simplified, but classification accuracy decreases due to bent and varying length chromosomes
Solution Approach 1:
The patent applies preprocessing operations including straightening bent chromosomes and normalizing their lengths before feeding them to the deep learning model. This preliminary action ensures that all chromosomes are presented in a standardized format, allowing the model to focus on classification features rather than being confounded by variations in shape and size, thereby improving accuracy.
Solution Approach 2:
The patent transforms chromosome images by changing key parameters such as orientation (straightening) and scale (normalizing length). These parameter changes convert variable-length, bent chromosomes into standardized representations, enabling the deep learning model to achieve higher classification accuracy without increasing model complexity.
4Measurement precision
If human experts perform chromosome segmentation and classification, then accuracy is maintained, but the process becomes laborious and expensive
Solution Approach 1:
The patent implements a hybrid system where the automated deep learning model performs the majority of segmentation and classification tasks independently. Human experts are only involved when the system identifies uncertain or complex cases that require manual review. This self-service approach with selective expert intervention significantly reduces labor costs while maintaining high accuracy.
Solution Approach 2:
The patent incorporates feedback mechanisms where expert corrections and manual annotations are used to retrain and improve the deep learning model. This feedback loop allows the system to learn from expert decisions, progressively reducing the need for manual intervention while maintaining or improving accuracy over time, thereby reducing long-term costs.
Data Source
Figure 1
Figure 2A
Figure 2B
AI summary
The most challenging problems in karyotyping are segmentation and classification of overlapping chromosomes in metaphase spread images. Often chromosomes are bent in different directions with varying degrees of bend. Tediousness and time consuming nature of the effort for ground truth creation makes it difficult to scale the ground truth for training phase. The present disclosure provides an end-to-end solution that reduces the cognitive burden of segmenting and karyotyping chromosomes. Dependency on experts is reduced by employing crowdsourcing while simultaneously addressing the issues associated with crowdsourcing. Identified segments through crowdsourcing are pre-processed to improve classification achieved by employing deep convolutional network (CNN).