Computer Vision Deep Learning for Rapid Diagnostic Test Analysis
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
The challenge lies in accurately capturing and interpreting results from lateral flow-based In-Vitro Rapid Diagnostic Tests (IV-RDTs), particularly in self-testing scenarios, due to difficulties in image acquisition and analysis, including optimal camera positioning, shadow and reflection issues, and the need for specific templates for each test device format.
Innovation Solution
A computer vision deep learning technique is employed, where a client device with a camera guides the user to capture images of the test region using augmented reality instructions, and a deep learning module analyzes these images to identify device features and classify test results based on visual indicators, determining the presence or level of biomarkers.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If manual image capture is used for RDT results, then user operation flexibility is maintained, but positioning accuracy and image quality deteriorate due to difficulty in capturing optimal test region images
Solution Approach 1:
The patent replaces manual mechanical image capture operations with an automated computer vision system. The deep learning model automatically detects and captures the test region by analyzing video streams from the camera, substituting the manual positioning and framing process with algorithmic image processing and automated capture based on detected test region coordinates.
Solution Approach 2:
The system enables self-service by allowing the test device to guide its own image capture. The computer vision system automatically identifies the test region, determines optimal capture parameters, and triggers image acquisition without requiring user expertise in positioning or framing, making the process self-correcting and autonomous.
2Measurement precision
If multiple specific templates are created for different test device formats, then recognition accuracy for each device type is improved, but system complexity increases due to need for multiple templates
Solution Approach 1:
The patent implements a universal deep learning model that can recognize multiple test device formats using a single system. The model is trained on diverse device types and can adaptively identify features across different formats (lateral flow assays, immunochromatographic tests, etc.), eliminating the need for separate templates for each device type while maintaining high recognition accuracy.
Solution Approach 2:
The system uses parameter changes by adjusting the deep learning model's detection parameters dynamically based on the input image characteristics. Rather than using fixed templates, the model learns to adapt its feature detection parameters (such as region boundaries, color thresholds, and pattern recognition criteria) to match the specific device format being analyzed.
3Productivity
If automated camera positioning is implemented, then image acquisition speed is improved, but system complexity increases due to need for imaging directing commands
Solution Approach 1:
The patent merges the computer vision analysis function with the camera control function into an integrated system. The same deep learning model that detects the test region also generates the imaging directing commands, combining image analysis and camera positioning tasks into a unified workflow that reduces overall system complexity while maintaining automated positioning capability.
Data Source
AI summary
A method and system include receiving a medical test request for a patient to perform a medical test to measure a biomarker using a rapid diagnostic test (RDT). The patient is instructed to perform the RDT. A first image stream of the RDT, including a test region displaying a visual indicator, is received from a camera coupled to the client device. A computer vision technique of a deep learning module is used to identify device features in the first image stream. A second image stream including the test region is received from the camera and applied to the deep learning module to: identify and classify the device features of the test region. A presence, a level, or both of the biomarker are determined from an RDT result based on the visual indicator in a second image.