Quality-Aware Retinal Vessel Classification in Low-Quality Fundus Images
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Low quality fundus images affect the accuracy of retinal artery or vein classification in existing deep learning models, leading to suboptimal processing effects.
Innovation Solution
A method that enhances image quality and predicts vessel locations using a first image processing model, trained with a loss function based on high-quality images and labeled information, to generate a second model capable of accurate vessel classification.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Ease of operation
If a pre-trained end-to-end deep learning model is used for retinal A/V classification, then the model can process fundus images directly, but the classification accuracy deteriorates when image quality is low
Solution Approach 1:
The patent segments the original end-to-end classification model into two separate models: an image quality assessment model and a vessel classification model. This segmentation allows the system to first evaluate image quality and then route low-quality images through preprocessing steps, thereby maintaining high classification accuracy while preserving direct processing capability for quality images.
2Ease of manufacture
If the deep learning model is trained using standard fundus image samples, then the training process is simple, but the model performance deteriorates on low quality images
Solution Approach 1:
The patent introduces a preliminary image quality assessment step before the main classification process. By pre-evaluating image quality and applying necessary preprocessing only when needed, the system maintains training simplicity while significantly improving model performance on low-quality images through targeted data preparation.
3Productivity
If no image quality consideration is incorporated during training, then the training process is fast and simple, but the classification accuracy on low quality images is low
Solution Approach 1:
The patent introduces an image quality assessment model as an intermediary component that mediates between the input image and the classification model. This intermediary evaluates image quality and determines appropriate preprocessing steps, enabling the system to maintain fast training speeds while achieving high classification accuracy on low-quality images through quality-aware processing.
Data Source
AI summary
This present disclosure relates to the technical field of artificial intelligence, and provides a vessel image classification method and apparatus, a device, and a storage medium. The method includes: inputting a first vessel image sample into a first image processing model, and obtaining a predicted enhanced image and predicted vessel location information; and training the first image processing model based on a second vessel image sample, vessel location labeling information, the predicted enhanced image, and the predicted vessel location information. In the above solution, the impact of image quality on the vessel classification is considered during training of the vessel classification model, so that an end-to-end vessel classification model subsequently generated based on the trained first image processing model can realize a higher classification accuracy for a low quality vessel image, thereby improving the accuracy of classifying vessels in the vessel image by artificial intelligence.