Particle Path Imaging With AI for Fast Motility Characterization
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current methods for characterizing motile particles, such as spermatozoa, are computationally costly and time-consuming, especially when dealing with high particle counts, exceeding tens of seconds per particle for thousands of particles due to N^3 log(N) complexity.
Innovation Solution
A method involving defocused or lensless imaging combined with a supervised-learning artificial-intelligence algorithm, such as a convolutional neural network, to process images of particles, forming a path image and computing average movement parameters, reduces computation time by combining multiple images to enhance detection and characterization.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If traditional tracking algorithms are used to characterize moving particles, then measurement precision is improved, but computation time increases significantly (N^3 log(N) complexity)
Solution Approach 1:
The patent replaces traditional mechanical tracking algorithms with a deep learning-based system. A convolutional neural network is trained to directly predict particle trajectories and characteristics from images, eliminating the need for complex pixel-by-pixel tracking computations. This substitution of computational methodology reduces complexity from N^3 log(N) to much faster deep learning inference time.
Solution Approach 2:
The patent creates a computational model (neural network) that learns from training data to replicate particle tracking results. Instead of performing exhaustive tracking computations on each new image, the system uses a pre-trained model that has copied the essential tracking logic from training examples, enabling rapid prediction of particle paths and characteristics without re-computing everything from scratch.
2Measurement precision
If the number of particles to be characterized is increased, then measurement precision is improved, but device complexity and computation time increase
Solution Approach 1:
The patent replaces traditional iterative tracking algorithms with a deep learning model that can handle large numbers of particles simultaneously. The convolutional neural network processes entire images or regions of interest in a single forward pass, capable of tracking thousands of particles without the N^3 log(N) complexity that plagues traditional methods. This allows high-precision characterization of large particle populations with manageable computational resources.
Applied Scientific Principles
This section explains which scientific principles are used to turn an abstract innovation direction into a practical engineering solution.
Function Achieved in This Case
The method significantly decreases analysis time while maintaining accurate characterization of particle movements, suitable for high particle counts, with correlation coefficients close to 1 and determination coefficients R^2 above 0.9 for various characteristics.
Implementation Method 1
acquiring at least one image of the sample during an acquisition period, using an image sensor defining a field of view
Implementation Method 2
an optical system lies between the sample and the image sensor, the optical system defining an object plane and an image plane
Implementation Method 3
the object plane is offset with respect to the sample plane by an object defocusing distance and/or the image plane is offset with respect to the sample plane by an image defocusing distance
Implementation Method 4
employing the path image resulting from b) as input image of a detection algorithm programmed to detect the particles and of a supervised-learning artificial-intelligence algorithm programmed to compute at least one average movement parameter
Data Source
AI summary
Method for characterizing at least one moving particle (10i) in a sample (10), the method comprising:a) acquiring at least one image (I, In) of the sample during an acquisition period, using an image sensor (20) defining a field of view, the acquisition period comprising various acquisition times (tn);b) using the image or each image resulting from a), forming a path image (I) showing the particles of the sample, in the field of view, at the various acquisition times;c) employing the path image resulting from b) as input image of a detection algorithm programmed to detect particles and of a supervised-learning artificial-intelligence algorithm programmed to compute at least one average movement parameter for various detected particles.