Microscopy Training Data Acquisition via Dynamic Segmentation
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
The acquisition of training data for machine learning models in microscopy is time-consuming and strains samples, as existing methods require capturing input and output images in both contrasts, often treating the entire sample, which can miss temporary morphological changes and increase sample damage.
Innovation Solution
A dynamic sample-and context-dependent method for acquiring training data, where images are analyzed based on predetermined criteria to determine optimal acquisition parameters, reducing unnecessary image capture and sample strain, allowing for context-dependent decision-making on data acquisition, including relevant image areas, time, and microscope settings.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If training data is acquired by capturing images in both contrasts for the entire sample, then sufficient training data is obtained, but sample strain increases and time consumption increases
Solution Approach 1:
The sample is divided into regions of interest based on preliminary analysis. Only these segmented regions are captured in both contrasts for training data acquisition, while other regions are excluded. This segmentation approach maintains data quality for relevant areas while significantly reducing the total area exposed to staining and imaging, thereby reducing sample strain.
Solution Approach 2:
Different quality requirements are applied to different regions of the sample. Regions containing relevant structures for the specific examination task are identified and assigned high priority for dual-contrast capture, while other regions are either captured in single contrast or excluded entirely. This local quality differentiation ensures sufficient training data for critical areas without unnecessarily straining the entire sample.
2Reliability
If training data is acquired by capturing images in both contrasts for the entire sample, then sufficient training data is obtained, but acquisition time increases
Solution Approach 1:
The sample area is segmented into regions of interest that require dual-contrast capture and other areas that can use single-contrast or pre-acquired images. This segmentation reduces the total pixel count requiring expensive dual-contrast acquisition, thereby significantly reducing acquisition time while maintaining data quality for the relevant regions.
Solution Approach 2:
Instead of applying dual-contrast capture uniformly across the entire sample (excessive action), the method applies it only partially to regions where it is truly necessary for training (partial action). This selective approach eliminates redundant acquisitions in areas that do not contribute meaningfully to the training objective, reducing overall acquisition time.
3Loss of time
If pre-trained models are used, then acquisition time is reduced, but adaptability to specific sample types decreases
Solution Approach 1:
A pre-trained model is used as a preliminary step to quickly analyze the sample and identify regions of interest before the main training data acquisition. This preliminary action leverages the speed of pre-trained models while enabling subsequent targeted acquisition that adapts to the specific sample type, thus combining time efficiency with adaptability.
Solution Approach 2:
The system dynamically adjusts the training data acquisition strategy based on the specific sample characteristics. Pre-trained models provide initial guidance, but the acquisition parameters (such as which regions to capture in both contrasts) are dynamically adapted to the specific sample type and examination task, ensuring both efficiency and adaptability.
Data Source
AI summary
In a method, a device and a computer program product for acquiring images for training data to train a statistical model by machine learning for image processing in microscopy, the training data is made up of pairs of input images and output images from image processing. The method includes acquiring at least one image, analyzing the at least one image according to predetermined criteria, determining acquisition parameters for the acquisition of output images on the basis of the analysis results, and acquiring output images on the basis of the determined acquisition parameters.