Virus-Induced CPE Detection in Unstained Live-Cell Images
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing methods for detecting virus-induced cytopathic effect (CPE) in cells are manual, time-consuming, and lack virus specificity, making them unsuitable for clinical applications and requiring chemical stains that interfere with live cell imaging.
Innovation Solution
A computer-implemented method using light microscopy and a trained convolutional neural network (CNN), specifically EfficientNet-BO, to label images of cells with or without CPE, allowing for virus-type specific detection without stains, and determining the infectious dose (TCID50) automatically.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If manual annotation methods are used for CPE detection, then virus infection can be identified, but the process is time-consuming and slow with several days readout
Solution Approach 1:
The patent replaces manual mechanical annotation processes with an automated machine learning system. A trained machine learning model automatically analyzes microscopy images to detect CPE, substituting the manual visual inspection process with computational analysis that delivers results in minutes rather than days, while maintaining or improving detection accuracy
Solution Approach 2:
The system enables self-service automation where the machine learning model independently performs CPE detection without requiring manual intervention. The model is trained once on annotated data and then autonomously analyzes subsequent images, eliminating the need for continuous manual annotation while providing rapid results
2Measurement precision
If chemical stains like crystal violet are used for CPE detection, then infected cells can be visualized, but live cell imaging is interfered with and virus specificity is lost
Solution Approach 1:
The patent extracts and removes the need for chemical stains from the detection process. By training the machine learning model to recognize CPE features directly in unstained live cell images, the harmful chemical staining step is completely eliminated, allowing continuous monitoring of live cells without interference while maintaining detection capability
Solution Approach 2:
The machine learning model serves as an intermediary that enables detection without chemical stains. Instead of using stains to make infected cells visible, the model acts as a computational mediator that identifies CPE features in natural light microscopy images of live cells, eliminating the need for harmful chemical intermediaries
3Measurement precision
If traditional TCID50 assays are performed manually, then infectious dose can be determined, but automation is lacking and virus specificity is not achieved
Solution Approach 1:
The patent replaces manual TCID50 assay procedures with an automated machine learning-based system. The model automatically analyzes images from serial dilution experiments, calculates infectious doses, and provides virus-type specific results, substituting manual counting and calculation processes with automated computational analysis
Solution Approach 2:
The machine learning model provides universal automation capability that handles multiple functions: it detects CPE, determines infectious doses, and identifies virus types across different experiments. This single automated system replaces multiple manual operations, achieving high extent of automation while maintaining measurement precision
Data Source
Figure 1~2
Figure 3~4C
Figure 5A~5D
AI summary
The present invention relates to a computer-implemented method (100) for training a machine learning model to label images of a plurality of cells as having or not having virus-induced cytopathic effect, comprising the steps of: a. Obtaining a training set including a plurality of images of virus-infected cells and uninfected cells, wherein the training set is annotated with labels indicating the presence or absence of virus-induced cytopathic effect for each image, and wherein the training set includes images of at least a set of cells infected by a one virus type and a set of cells infected by a second virus type; b. Training the machine learning model on the training set to predict whether an image of a plurality of cells contains virus-induced cytopathic effect or not, based on features extracted from the images. The present invention relates also to a computer-implemented method for labelling an image of a plurality of cells as having or not having virus-induced cytopathic effect