AI Microdroplet Segmentation for Accurate Digital Assays
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing digital assay methods, such as digital PCR, face challenges in determining thresholds for micro-sized droplets, suffer from optical coherence issues, and require separate fluorescence readers, leading to user bias and reduced accuracy due to background variation and noise spikes.
Innovation Solution
A digital assay method using an artificial intelligence-based prediction model that segments microdroplets and applies a deep learning algorithm to count positive, negative, and atypical microdroplets, enabling accurate quantitative analysis without user intervention.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If traditional threshold-based segmentation methods are used for microdroplet detection, then the device complexity is low, but the measurement precision deteriorates due to difficulty in determining thresholds and user bias
Solution Approach 1:
The patent replaces traditional mechanical/optical threshold-based segmentation methods with an artificial intelligence-based prediction model. The model automatically identifies and segments microdroplets in images, eliminating the need for manual threshold determination and reducing user bias. This substitution of AI algorithms for traditional image processing methods improves measurement precision while managing system complexity through software-based solutions.
2Measurement precision
If separate fluorescence readers are used for digital PCR, then the measurement precision can be maintained, but the device complexity increases and reliability decreases due to optical coherence issues
Solution Approach 1:
The patent integrates the imaging and detection functions into a unified system. The AI-based prediction model processes images captured by the imaging unit to identify microdroplets and quantify targets, eliminating the need for separate fluorescence readers. This merging of functions reduces device complexity and avoids optical coherence issues while maintaining quantification accuracy through intelligent image analysis.
3Measurement precision
If adaptive threshold methods are used to improve analysis accuracy, then the measurement precision improves, but the ease of operation deteriorates due to requiring user intervention for threshold determination
Solution Approach 1:
The AI-based prediction model performs automatic threshold determination and microdroplet segmentation without requiring user intervention. The model learns optimal segmentation parameters from training data and applies them automatically to new images, enabling the system to serve itself rather than requiring manual threshold adjustment. This self-service capability improves ease of operation while maintaining high analysis accuracy.
4Reliability
If traditional image processing methods are used, then the device complexity is low, but the reliability deteriorates due to background variation and noise spikes affecting threshold determination
Solution Approach 1:
The patent replaces traditional noise-sensitive thresholding algorithms with an AI-based prediction model that is robust to background variation and noise. The model learns to distinguish true microdroplet signals from background noise and variations during training, enabling reliable detection even in challenging imaging conditions. This substitution improves result reliability while managing system complexity through sophisticated software processing.
Data Source
Figure 1
Figure 2A
Figure 2B
AI summary
Provided are a device for a digital assay of targets according to an exemplary embodiment of the present disclosure and a method using the same. The digital assay method of targets according to the exemplary embodiment of the present disclosure includes acquiring an image for a plurality of microdroplets, predicting at least one region based on the image for the plurality of microdroplets using an artificial neural network-based prediction model configured to segment at least one region among positive microdroplets, negative microdroplets, and atypical microdroplets, with the image for the plurality of microdroplets as an input, determining a number for the plurality of microdroplets based on the at least one region, and providing quantitative data of targets based on the number for the plurality of microdroplets.