CNN Region Proposal Training for HIL Specimen Characterization

Resolve Bottlenecks,
Find Innovative Solutions
Generate Solutions

Solution Overview

Problem

Existing model-based systems for characterizing bio-fluid specimens are labor-intensive due to the need for manual graphical annotation during training, making them inefficient and time-consuming.

Innovation Solution

A method and apparatus that utilize a neural network, specifically a convolutional neural network (CNN), to automatically generate region proposals for serum or plasma portions in bio-fluid specimens without manual input, using captured images and ground truth labels to train the network efficiently.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Measurement precision

If manual graphical annotation is used during training of model-based systems, then the system can characterize bio-fluid specimens, but the training process becomes labor-intensive and time-consuming

Engineering Contradiction:
Improvespecimen characterization accuracyVSAvoidtraining time
Core Design Contradiction:
Measurement precisionVSLoss of time

Solution Approach 1:

The system performs self-annotation by automatically generating region proposals for serum and plasma portions without requiring manual graphical annotation. The neural network learns to identify and segment these regions autonomously during training, eliminating the need for labor-intensive manual labeling while maintaining characterization accuracy

Inventive Principle:
Principle #25Self-service

Solution Approach 2:

The system generates region proposals as preliminary annotations during the training process itself, rather than requiring pre-computed manual annotations. This preliminary automated segmentation enables the neural network to learn specimen characterization directly from raw images, significantly reducing training preparation time

Inventive Principle:
Principle #10Preliminary action

2Reliability

If manual graphical annotation is used during training, then the model can be trained to identify serum or plasma portions, but the process becomes inefficient and labor-intensive

Engineering Contradiction:
Improvemodel training reliabilityVSAvoidtraining efficiency
Core Design Contradiction:
ReliabilityVSProductivity

Solution Approach 1:

The patent replaces the mechanical process of manual graphical annotation with an automated computational process. A neural network component automatically generates region proposals for serum and plasma portions, substituting human labor with machine-based segmentation that maintains reliability while dramatically improving training efficiency

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

Solution Approach 2:

The system introduces an intermediary automated region proposal generation mechanism between the raw input images and the neural network training process. This intermediary automatically creates the segmentation masks needed for training, eliminating the need for manual annotation while ensuring the reliability of the training data

Inventive Principle:
Principle #24Intermediary (Mediator)

Data Source

PatentEP3737930B1Methods and apparatus for bio-fluid specimen characterization using neural network having reduced training
Publication Date: 2025.11.05 SIEMENS HEALTHCARE DIAGNOSTICS INC
  • EP3737930B1 patent drawingFigure 1~2
  • EP3737930B1 patent drawingFigure 3A~3B
  • EP3737930B1 patent drawingFigure 3C~3D

AI summary

A method of training a neural network (Convolutional Neural Network - CNN) including reduced graphical annotation input is provided. The training method can be used to train a Testing CNN that can be used for determining Hemolysis (H), Icterus (I), and/or Lipemia (L), or Normal (N) of a serum or plasma portion of a test specimen. The training method includes capturing training images of multiple specimen containers including training specimens, generating region proposals of the serum or plasma portions of the training specimens; and selecting the best matches for the location, size and shape of the region proposals for the multiple training specimens. The obtained features (network and weights) from the training CNN can be used in a testing CNN. Quality check modules and testing apparatus adapted to carry out the training method, and characterization methods using a bounding box regressor are described, as are other aspects.