Deep Learning Phase Image Reconstruction from Brightfield Data

Resolve Bottlenecks,
Find Innovative Solutions
Generate Solutions

Solution Overview

Problem

Brightfield phase imaging techniques face challenges such as artifacts like dust particles, water droplets, and well borders, which lead to distorted phase images due to absorption effects not accounted for in the phase equation, and require substantial time for image generation, resulting in blurring at image edges and background issues.

Innovation Solution

A machine learning model is trained using simulated visual artifacts to reconstruct high-quality phase images from brightfield images, reducing the impact of artifacts and improving image clarity by learning to adapt to and compensate for visual disturbances, such as dust, water droplets, and well borders, within the training data set.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Productivity

If traditional phase imaging methods are used to process brightfield images, then phase information can be reconstructed, but the processing time is substantial (3-5 seconds per image) and artifacts are not effectively removed

Engineering Contradiction:
Improveimage generation speedVSAvoidimage quality with artifacts
Core Design Contradiction:
ProductivityVSReliability

Solution Approach 1:

The patent replaces traditional mechanical/optical phase imaging systems with a deep learning-based neural network system. The neural network is trained on simulated data to learn phase reconstruction and artifact removal simultaneously, substituting the physical phase imaging process with an computational model that achieves both speed and quality improvement.

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

Solution Approach 2:

The patent performs preliminary training of the neural network using simulated brightfield images with known ground truth phase images. This pre-training phase allows the system to learn artifact patterns and phase reconstruction mappings before actual imaging, enabling fast and accurate phase reconstruction without artifacts during operation.

Inventive Principle:
Principle #10Preliminary action

2Measurement precision

If the phase equation is applied to reconstruct phase images from brightfield images, then phase information is obtained, but absorption effects from artifacts cause nonlocal distortion in the phase solution

Engineering Contradiction:
Improvephase image accuracyVSAvoidartifact absorption effects
Core Design Contradiction:
Measurement precisionVSObject-affected harmful factors

Solution Approach 1:

The patent converts the harmful effect of artifacts into a beneficial training opportunity. By simulating various artifacts in the training data and corresponding ground truth phase images, the neural network learns to recognize and compensate for artifact absorption effects, transforming what would be sources of error into features that improve reconstruction accuracy.

Inventive Principle:
Principle #22Blessing in disguise (Convert harm into benefit)

Solution Approach 2:

The neural network acts as an intermediary between the raw brightfield images and the final phase reconstruction. It processes the input images through multiple layers that learn to separate true phase information from artifact-induced distortions, providing a cleaned phase solution that accounts for absorption effects without requiring direct application of the phase equation.

Inventive Principle:
Principle #24Intermediary (Mediator)

3Measurement precision

If multiple focal planes are imaged to enhance phase imaging, then image clarity is improved, but the complexity of processing multiple images increases computation time

Engineering Contradiction:
Improvephase image clarityVSAvoidprocessing time for multiple images
Core Design Contradiction:
Measurement precisionVSLoss of time

Solution Approach 1:

The patent merges the tasks of processing multiple focal planes, removing artifacts, and reconstructing phase information into a single unified neural network operation. The network accepts multiple focal plane images as input and simultaneously performs all necessary processing steps, eliminating the sequential processing time required by traditional methods.

Inventive Principle:
Principle #5Merging (Combining)

Solution Approach 2:

The patent employs a dynamic neural network architecture that can adaptively process variable numbers of focal planes. The network learns optimal processing strategies from training data with different focal plane configurations, enabling efficient handling of multiple images without fixed computational overhead.

Inventive Principle:
Principle #15Dynamics

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

The approach significantly reduces the time required to generate phase images, achieving enhanced image clarity and isolation of cells from background content, while effectively addressing the limitations of traditional methods by automatically removing or reducing the impact of visual artifacts.

Implementation Method 1

Due to the nature of optical refraction through the different elements included in the sample, a pair of brightfield images at different focal planes of a sample may be processed and combined to reconstruct a clear image of the sampled subject

Methodology Applied
Scientific EffectRefraction: Refraction

Data Source

PatentEP4035129B1Reconstructing phase images with deep learning
Publication Date: 2024.05.01 PERKINELMER CELLULAR TECH GERMANY GMBH
  • EP4035129B1 patent drawingFigure 1
  • EP4035129B1 patent drawingFigure 2
  • EP4035129B1 patent drawingFigure 3A

AI summary

Aspects relate to reconstructing phase images from brightfield images at multiple focal planes using machine learning techniques. A machine learning model may be trained using a training data set comprised of matched sets of images, each matched set of images comprising a plurality of brightfield images at different focal planes and, optionally, a corresponding ground truth phase image. An initial training data set may include images selected based on image views of a specimen that are substantially free of undesired visual artifacts such as dust. The brightfield images of the training data set can then be modified based on simulating at least one visual artifact, generating an enhanced training data set for use in training the model. Output of the machine learning model may be compared to the ground truth phase images to tram the model. The trained model may be used to generate phase images from input data sets.