CNN Cell Image Characterization for Single-Cell Verification

Resolve Bottlenecks,
Find Innovative Solutions
Generate Solutions

Solution Overview

Problem

Current methods for analyzing cell line development images in biopharmaceutical production are inconsistent and inefficient, leading to unreliable identification of single cells and potential misclassification, which can impact the reliability of clonally-derived cell lines.

Innovation Solution

An automated imaging workflow using machine-learning algorithms, including convolutional neural networks, for image characterization that applies digital image processing and deep learning to filter out bad images and accurately identify single cells with good morphology, replacing manual analysis.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Reliability

If manual image analysis is used to identify single cells, then flexibility and adaptability are maintained, but consistency and efficiency of analysis are poor

Engineering Contradiction:
Improveconsistency of image analysisVSAvoidautomation of image review
Core Design Contradiction:
ReliabilityVSExtent of automation

Solution Approach 1:

The patent replaces manual mechanical image review with an automated machine learning system. A trained machine learning model automatically analyzes cell culture plate images to identify single cells, replacing the manual visual inspection process. This substitution maintains reliability through consistent algorithmic application while achieving full automation of the review process.

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

2Productivity

If automated machine learning image analysis is implemented, then throughput and consistency are improved, but accuracy may deteriorate due to misclassification

Engineering Contradiction:
Improvethroughput of image analysisVSAvoidaccuracy of single cell identification
Core Design Contradiction:
ProductivityVSMeasurement precision

Solution Approach 1:

The patent implements a feedback mechanism where the machine learning model's predictions are reviewed and validated. The system provides confidence scores for each identification, and low-confidence predictions can be flagged for manual review or re-analysis. This feedback loop ensures high accuracy while maintaining automated throughput by allowing correction of potential misclassifications.

Inventive Principle:
Principle #23Feedback

Solution Approach 2:

The patent applies preliminary image processing and feature extraction before final classification. The machine learning model is pre-trained on labeled data to learn accurate cell identification patterns. This preliminary preparation of the model and data ensures high measurement precision is achieved before the actual high-throughput analysis begins.

Inventive Principle:
Principle #10Preliminary action

3Measurement precision

If comprehensive manual inspection of all images is performed, then accuracy is maintained, but time consumption and efficiency are poor

Engineering Contradiction:
Improveaccuracy of cell classificationVSAvoidtime for image review
Core Design Contradiction:
Measurement precisionVSLoss of time

Solution Approach 1:

The patent applies partial action by using the machine learning model to screen all images automatically, then applying comprehensive manual inspection only to a subset of images that require verification. This selective approach maintains high accuracy for critical cases while dramatically reducing overall time consumption compared to inspecting every image manually.

Inventive Principle:
Principle #16Partial or excessive action

Data Source

PatentEP4078441B1Cell line development image characterization with convolutional neural networks
Publication Date: 2026.01.28 GENENTECH INC
  • EP4078441B1 patent drawingFigure 1
  • EP4078441B1 patent drawingFigure 2
  • EP4078441B1 patent drawingFigure 3

AI summary

A method includes, by a computing system, receiving a querying image depicting a sampling area, processing the querying image using a single cluster detection model to identify one or more regions of the querying image depicting a cluster in the sampling area, processing the one or more regions using a cluster verification deep-learning model to determine whether each depicted cluster is a cell cluster, and determining that exactly one of the identified one or more regions depicts a cluster that is a cell cluster. The method further includes processing the region depicting the cell cluster using a morphology deep-learning model to determine that there is only one cell in the cell cluster and to determine that the morphology of the cell is acceptable.