Deep Learning Cell Detection Without Fluorescent Markers
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current methods for drug discovery screening require marker introduction into cells, which is invasive, costly, and not suitable for continuous observation, making it difficult to automatically measure cell positions and numbers from marker-free images.
Innovation Solution
An image processing method using deep learning models to generate pseudo-marked images from marker-free images, allowing for the accurate detection of cell positions and numbers without the need for marker introduction, by constructing a first learning model to associate images with and without markers, and a second learning model to determine the positions of specific parts.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If marker introduction is performed to enable automatic measurement of cell positions and numbers, then measurement precision is improved, but device complexity and experiment cost increase due to additional reagents and operations
Solution Approach 1:
The patent uses deep learning models to generate synthetic fluorescent marker images from bright-field images. The first learning model generates pseudo-fluorescent images that copy the appearance of actual fluorescent images, and the second learning model detects cell positions from these generated images, eliminating the need for physical marker introduction while achieving equivalent measurement precision
Solution Approach 2:
The patent replaces the physical-chemical process of marker introduction with an information processing approach using deep neural networks. Instead of chemically labeling cells with fluorescent markers, the system uses image processing algorithms to predict and detect cell positions directly from bright-field images, substituting mechanical/chemical operations with computational methods
2Extent of automation
If marker introduction is performed to enable automatic cell detection, then automation is improved, but ease of operation deteriorates due to additional experimental steps and reagent handling
Solution Approach 1:
The system copies the functional outcome of marker introduction (fluorescent signal generation) through computational means. The deep learning models generate synthetic fluorescent images that replicate the information provided by actual markers, allowing automated detection without the operational burden of physical marker introduction
Solution Approach 2:
The patent extracts and separates the function of marker introduction from the cell detection process. By using bright-field images alone and generating synthetic fluorescent signals through deep learning, the system removes the marker introduction step while preserving the essential detection capability, simplifying the overall experimental workflow
3Ease of operation
If marker-free imaging is used to simplify operations and reduce cost, then ease of operation and cost are improved, but measurement precision deteriorates due to difficulty in automatically identifying cell positions
Solution Approach 1:
The patent introduces an intermediary computational process (deep learning-based image generation) between the bright-field image and cell position detection. The first learning model acts as an intermediary that converts bright-field images into synthetic fluorescent images, and the second learning model uses these intermediates to achieve precise position detection, thereby maintaining measurement precision while using simpler marker-free imaging
Solution Approach 2:
The system substitutes the optical-chemical detection mechanism (fluorescent marker emission) with a computational detection mechanism (neural network-based position prediction). This allows the system to achieve the same measurement precision as fluorescent imaging but using only bright-field imaging combined with deep learning analysis
4Measurement precision
If marker introduction is performed to enable continuous cell observation, then measurement capability is improved, but reliability deteriorates due to cell modification and potential death
Solution Approach 1:
The patent creates a virtual copy of the fluorescent marker signal through deep learning image generation. Instead of physically modifying cells with markers that may harm viability, the system generates synthetic fluorescent images from bright-field images, preserving cell integrity while enabling continuous observation and measurement
Solution Approach 2:
The patent extracts the essential measurement information (cell position and number) from bright-field images alone, removing the need for marker introduction that compromises cell viability. This allows continuous, long-term observation of live cells without the harmful effects of chemical labeling
Data Source
Figure 1
Figure 2~3
Figure 4
AI summary
An image processing method of the invention is, to enable automatically measure a position and a number of cells included in an image even if a marker is not used, configured to input a test image to a first learning model (Step S102), input an output image to a second learning model (Step S104) and output an output image as a result image in which the position of the detected part is indicated by a representative point (Step S105). The first learning model is constructed by deep learning using teacher data associating a first image in which a marker is expressed and a second image in which the marker is not expressed, the first and the second images being captured to include a same cell. The second learning model is constructed by deep learning using teacher data associating a third image, which is captured to include a cell and in which the marker is expressed, and information representing a position of the representative point included in the third image.