Automated Classifier Training via Spectral Unmixing

Resolve Bottlenecks,
Find Innovative SolutionsGenerate Solutions

Solution Overview

Problem

Automated machine-based classifiers for sample imaging and classification are time-consuming and prone to operator error due to the need for manual selection of regions of interest and the limitations of available stains, which can lead to inaccurate training and analysis.

Innovation Solution

A method involving the application of spectrally distinguishable stains to samples, followed by spectral unmixing to obtain component images, which are used to train automated classifiers to identify specific regions of interest, reducing manual labor and improving accuracy by leveraging distinct spectral contributions from different stains or autofluorescence.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Reliability

If manual classification is used to train automated classifiers, then the classifier can be trained with available stains, but the process is time-consuming and prone to operator error

Engineering Contradiction:

Improveclassification accuracyVSAvoidtraining time

Core Design Contradiction:

ReliabilityVSLoss of time

Solution Approach 1:

The system performs self-training by automatically identifying regions of interest through spectral unmixing and component image analysis, eliminating the need for manual operator intervention in the training process while maintaining high accuracy

Inventive Principle:

Principle #25Self-service

Solution Approach 2:

The patent replaces manual mechanical classification with automated image processing and spectral analysis algorithms, substituting human operator actions with computational methods that are both faster and more consistent

Inventive Principle:

Principle #28Mechanics substitution (Replace mechanical system)

2Adaptability or versatility

If manual selection of regions of interest is performed, then training can be done with limited stains, but operator error increases and consistency decreases

Engineering Contradiction:

Improvestain flexibilityVSAvoidregion identification accuracy

Core Design Contradiction:

Adaptability or versatilityVSMeasurement precision

Solution Approach 1:

The patent utilizes spectral unmixing to separate and identify different stains based on their unique spectral signatures, allowing automatic differentiation of regions of interest without relying on manual interpretation of color variations

Inventive Principle:

Principle #32Color changes

Solution Approach 2:

The system introduces component images as an intermediary representation that isolates spectral contributions from individual stains, enabling automated and consistent identification of regions of interest without direct manual intervention

Inventive Principle:

Principle #24Intermediary (Mediator)

3Extent of automation

If spectral unmixing and component images are used, then automated training accuracy improves, but the complexity of the imaging and analysis process increases

Engineering Contradiction:

Improveautomated classificationVSAvoidimaging system complexity

Core Design Contradiction:

Extent of automationVSDevice complexity

Solution Approach 1:

The patent segments the complex spectral information into separate component images, each representing the contribution of a specific stain, thereby simplifying the automated analysis process while maintaining high accuracy in region identification

Inventive Principle:

Principle #1Segmentation

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

This approach reduces the time and error associated with manual classification, enabling more accurate and efficient training of automated classifiers to identify various regions of interest in samples, even when universal stains are not available, and allows for the analysis of samples with varying types and sub-cellular components.

Implementation Method 1

applying a first stain to a first sample having a plurality of regions, where the first stain selectively binds to only a first subset of the regions of the first sample

Methodology Applied

Scientific EffectSpectral binding: Absorption Spectroscopy

Implementation Method 2

applying a second stain to the first sample, where the second stain binds to a second set of regions of the first sample

Methodology Applied

Scientific EffectSpectral binding: Absorption Spectroscopy

Implementation Method 3

analyzing the image to obtain a first component image corresponding substantially only to spectral contributions from the first stain, and a second component image corresponding substantially only to spectral contributions from the second stain

Methodology Applied

Scientific EffectSpectral unmixing: Absorption Spectroscopy

Data Source

PatentUS8379962B2Image classifier training

Publication Date: 2013.02.19 AKOYA BIOSCIENCES INC

AI summary

Methods are disclosed that include: (a) applying a first stain to a first sample having a plurality of regions, where the first stain selectively binds to only a first subset of the regions of the first sample; (b) applying a second stain to the first sample, where the second stain binds to a second set of regions of the first sample; (c) obtaining an image of the first sample, and analyzing the image to obtain a first component image corresponding substantially only to spectral contributions from the first stain, and a second component image corresponding substantially only to spectral contributions from the second stain; and (d) training a classifier to identify regions of a second sample based on information derived from the first and second component images, the identified regions corresponding to the first subset of regions of the first sample.