Multidimensional Event Classification via Density Clustering

Resolve Bottlenecks,
Find Innovative Solutions
Generate Solutions

Solution Overview

Problem

Current methods for automatic classification of events in multidimensional spaces are inefficient, particularly when dealing with large datasets from biomedical analyses, as they require prior knowledge, are subjective, and struggle with non-Gaussian distributions or unknown populations, leading to incomplete analysis and inaccuracies.

Innovation Solution

A computer-implemented method that clusters events by determining event density, connecting closest neighbors, calculating affinity between groups, and comparing with reference groups using dimensionality reduction and statistical criteria to identify populations without requiring prior knowledge of group numbers or distributions.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Measurement precision

If manual analysis methods are used to classify events in multidimensional space, then analysis precision can be maintained for small datasets, but productivity decreases dramatically as the number of parameters increases

Engineering Contradiction:
Improveclassification precisionVSAvoidanalysis speed
Core Design Contradiction:
Measurement precisionVSProductivity

Solution Approach 1:

The patent replaces manual mechanical analysis with an automated computational system that uses algorithms to cluster and classify events in multidimensional space, eliminating the need for manual inspection of numerous parameter combinations while maintaining classification precision

Inventive Principle:
Principle #28Mechanics substitution (Replace mechanical system)

Solution Approach 2:

The patent transforms the analysis approach by changing from manual parameter-by-parameter examination to automated multivariate analysis, where the system automatically adjusts and optimizes parameter combinations to identify populations without manual intervention

Inventive Principle:
Principle #35Parameter changes

2Productivity

If automatic clustering methods based on finite mixture models are used, then productivity increases, but reliability decreases because the methods require prior knowledge of group numbers and thresholds

Engineering Contradiction:
Improveanalysis speedVSAvoidanalysis objectivity
Core Design Contradiction:
ProductivityVSReliability

Solution Approach 1:

The patent implements a self-service classification system where the algorithm automatically determines the number of groups and thresholds by analyzing the data distribution itself, without requiring pre-programmed knowledge of population characteristics or manual setting of classification criteria

Inventive Principle:
Principle #25Self-service

Solution Approach 2:

The patent introduces dynamic adaptation where the classification system automatically adjusts its parameters and thresholds based on the specific dataset being analyzed, allowing it to adapt to different population distributions and characteristics without requiring manual reconfiguration

Inventive Principle:
Principle #15Dynamics

3Productivity

If conventional automatic classification methods are used, then productivity improves, but measurement precision deteriorates when dealing with non-Gaussian distributions or unknown populations

Engineering Contradiction:
Improveautomation levelVSAvoidclassification accuracy
Core Design Contradiction:
ProductivityVSMeasurement precision

Solution Approach 1:

The patent changes the mathematical approach from assuming Gaussian distributions to using non-parametric methods that can handle any data distribution, allowing accurate classification of unknown populations without requiring the data to follow specific statistical patterns

Inventive Principle:
Principle #35Parameter changes

Solution Approach 2:

The patent segments the classification process into distinct stages: density estimation, cluster formation, and validation, where each stage independently handles different aspects of the data, allowing the system to process complex non-Gaussian distributions through systematic decomposition of the classification task

Inventive Principle:
Principle #1Segmentation

4Measurement precision

If manual step-by-step classification is performed for each population, then measurement precision is maintained, but device complexity increases as the number of steps increases with the number of parameters

Engineering Contradiction:
Improveclassification accuracyVSAvoidnumber of analysis steps
Core Design Contradiction:
Measurement precisionVSDevice complexity

Solution Approach 1:

The patent merges multiple manual classification steps into a single integrated automated algorithm that simultaneously handles multiple parameters and populations, consolidating what would otherwise require numerous separate analysis steps into one cohesive computational process

Inventive Principle:
Principle #5Merging (Combining)

Data Source

PatentEP3340107B9Method of digital information classification
Publication Date: 2021.07.21 CYTOGNOS
  • EP3340107B9 patent drawingFigure 1~2
  • EP3340107B9 patent drawingFigure 3~4
  • EP3340107B9 patent drawingFigure 5(a)~6

AI summary

The invention relates to a computer-implemented method for classifying events present in a sample, wherein each event is characterized by a multidimensional set of parameters obtained by means of hardware and/or software, wherein the values of the parameters associated with each event define the position coordinates of said event in a multidimensional space. Said method comprises the following stages: a) clustering the events in groups, b) checking if within each formed group there is a connection between events exceeding a maximum distance threshold, c) calculating the affinity between each pair of previously generated sample groups, d) comparing each sample group with at least one reference group stored in at least one database, e) classifying the sample groups based on the comparisons with the reference groups.