Super-Resolution Imaging Accuracy via AI Artifact Classification
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current super-resolution imaging techniques face challenges in accurately determining which artifacts at low resolution magnification are suitable for generating high-quality super-resolution images, leading to potential inaccuracies and the need for additional high-resolution scanning.
Innovation Solution
A method and system that utilize artificial intelligence, including machine learning and neural networks, to assess the suitability of artifacts for super-resolution imaging by comparing low-resolution images to known artifacts, determining an image grade, and selectively employing higher resolution scanning for problematic areas, thereby improving the accuracy of super-resolution image generation.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If super-resolution imaging is performed on all detected artifacts, then image accuracy is improved, but processing time and computational resources increase significantly
Solution Approach 1:
The system performs preliminary classification of artifacts into suitable and unsuitable categories before super-resolution processing. This preliminary action identifies which artifacts require high-resolution imaging, preventing unnecessary processing of unsuitable artifacts and reducing overall processing time while maintaining accuracy for relevant targets.
Solution Approach 2:
The system applies different processing qualities to different artifacts based on their suitability characteristics. Suitable artifacts receive full super-resolution processing for maximum accuracy, while unsuitable artifacts are handled differently (e.g., rejected or processed with lower resources), optimizing the allocation of computational resources to where they provide the most value.
2Measurement precision
If higher resolution objectives are used, then optical resolution is improved, but scanning speed decreases
Solution Approach 1:
The system performs preliminary identification of suitable artifacts using lower resolution scanning, then selectively applies higher resolution objectives only to those identified artifacts. This preliminary sorting enables the system to maintain fast low-resolution scanning for most areas while achieving high resolution only where necessary, balancing scanning speed and optical resolution.
Solution Approach 2:
Instead of applying high-resolution scanning to the entire specimen, the system applies it partially only to suitable artifacts that meet specific criteria. This partial action approach achieves the necessary optical resolution for critical targets without the excessive time cost of full-specimen high-resolution scanning.
3Measurement precision
If AI models are trained on all low-resolution images, then super-resolution accuracy is improved, but training time and computational resources increase
Solution Approach 1:
The system extracts and selects only suitable artifacts from the full dataset for inclusion in training sets. By filtering out unsuitable artifacts, the training process focuses computational resources on high-quality examples that will improve super-resolution accuracy, reducing training time while maintaining or enhancing model performance.
Solution Approach 2:
The training process applies different data quality standards to different artifacts. Suitable artifacts with clear characteristics are prioritized for training, while unsuitable artifacts are excluded. This local quality approach ensures training resources are concentrated on examples that provide the most learning value for super-resolution accuracy.
Data Source
Figure 1
Figure 2A
Figure 2B
AI summary
Systems, methods, and computer-readable media for feedback on and improving the accuracy of super-resolution imaging. In some embodiments, a low resolution image of a specimen can be obtained using a low resolution objective of a microscopy inspection system. A super-resolution image of at least a portion of the specimen can be generated from the low resolution image of the specimen using a super-resolution image simulation. Subsequently, an accuracy assessment of the super-resolution image can be identified based on one or more degrees of equivalence between the super-resolution image and one or more actually scanned high resolution images of at least a portion of one or more related specimens identified using a simulated image classifier. Based on the accuracy assessment of the super-resolution image, it can be determined whether to further process the super-resolution image. The super-resolution image can be further processed if it is determined to further process the super-resolution image.