Microscopy ROI Classification With Confidence-Guided Re-Analysis

Resolve Bottlenecks,
Find Innovative Solutions
Generate Solutions

Solution Overview

Problem

Conventional methods for classifying microscopic components of physical samples are slow, require significant human effort for training data creation, and lack reproducibility across different microscopy systems, leading to inefficiencies in high-volume analysis applications.

Innovation Solution

A machine-learning-based system that autonomously or semi-autonomously identifies regions-of-interest (ROIs) in sample images, applies trained models for initial classification, and re-analyzes using a second mode when confidence is low, leveraging different analysis modes to enhance accuracy without human intervention.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Productivity

If conventional methods are used for classifying microscopic components, then human effort for training data creation is required, but analysis speed is slow and productivity is low

Engineering Contradiction:
Improveanalysis speedVSAvoidtime for training data creation
Core Design Contradiction:
ProductivityVSLoss of time

Solution Approach 1:

The system performs self-training by automatically generating training data from its own analysis results. The classification module generates initial classifications, and these results are fed back to train and improve the classification model without requiring external human intervention for data creation, enabling continuous self-improvement and high-speed analysis

Inventive Principle:
Principle #25Self-service

Solution Approach 2:

The system performs preliminary classification using a fast classification module to generate initial classifications and confidence scores before committing to final classifications. This preliminary action allows the system to quickly filter and prioritize samples that need further analysis, significantly improving overall analysis speed

Inventive Principle:
Principle #10Preliminary action

2Measurement precision

If conventional methods are used for classification, then simple analysis processes are used, but measurement precision and classification accuracy are insufficient

Engineering Contradiction:
Improveclassification accuracyVSAvoidanalysis process complexity
Core Design Contradiction:
Measurement precisionVSDevice complexity

Solution Approach 1:

The classification system is segmented into multiple specialized modules: a classification module for generating initial classifications and confidence scores, and a second classification module for handling low-confidence cases. This segmentation allows each module to be optimized for its specific function, improving overall classification accuracy while managing complexity through modular design

Inventive Principle:
Principle #1Segmentation

Solution Approach 2:

The system implements feedback loops where classification results and confidence scores are fed back into the training process. Low-confidence classifications trigger re-analysis or human review, and all results are used to continuously retrain and improve the classification models, progressively enhancing measurement precision

Inventive Principle:
Principle #23Feedback

3Reliability

If conventional methods are used for training, then significant human effort is required, but reproducibility across different microscopy systems is poor

Engineering Contradiction:
ImprovereproducibilityVSAvoidhuman effort for training
Core Design Contradiction:
ReliabilityVSEase of operation

Solution Approach 1:

The system uses universal training data generated from analysis results that can be applied across different microscopy systems and analysis modes. The self-training mechanism creates reproducible models that work consistently across various systems without requiring system-specific manual training, enhancing both reproducibility and ease of operation

Inventive Principle:
Principle #6Universality (Multi-functionality)

Solution Approach 2:

The system replaces manual human training operations with automated machine learning algorithms. The classification module automatically generates training data and retrains models without human intervention, eliminating the variability and effort associated with manual training while ensuring consistent, reproducible results across different systems

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

4Productivity

If fast analysis modes are used, then productivity increases, but measurement precision and classification reliability decrease

Engineering Contradiction:
Improveanalysis throughputVSAvoidclassification reliability
Core Design Contradiction:
ProductivityVSMeasurement precision

Solution Approach 1:

The system dynamically adjusts its analysis depth based on confidence scores. High-confidence classifications from fast analysis are accepted immediately, while low-confidence cases trigger additional analysis passes or human review. This dynamic approach maintains high throughput for clear cases while ensuring accuracy for ambiguous cases

Inventive Principle:
Principle #15Dynamics

Solution Approach 2:

The system performs partial re-analysis only on low-confidence classifications rather than re-analyzing all samples. By applying excessive analysis only where needed (low-confidence cases), the system maintains high overall productivity while ensuring reliability for critical classifications

Inventive Principle:
Principle #16Partial or excessive action

Data Source

PatentEP4648020A1Classifying microscopic components of physical samples
Publication Date: 2025.11.12 FEI CO
  • EP4648020A1 patent drawingFigure 1~2
  • EP4648020A1 patent drawingFigure 3
  • EP4648020A1 patent drawingFigure 4

AI summary

Disclosed herein are systems for classifying microscopic components of physical samples, as well as related methods, computing devices, and computer-readable media. For example, in some embodiments, a method for classifying microscopic components of a physical sample may include: generating a set of regions-of-interest (ROIs) in an image representative of the physical sample, wherein the image is generated by a microscopy system using a first analysis mode; generating an initial classification for an ROI by applying a trained machine-learning model to at least the portion of the image associated with the ROI; generating a confidence score associated with the initial classification; and when the confidence score for an initial classification of an ROI does not satisfy a set of confidence criteria, causing the microscopy system to re-analyze at least the portion of the sample associated with the ROI using a second analysis mode different than the first analysis mode.