Fourier Ptychographic Refocusing for Uniform-Focus Pathology Deep Learning
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Pathology analysis, particularly in histologic/cytologic images, faces challenges with focus quality variability and inefficiencies in acquiring uniformly focused images, leading to suboptimal performance of deep learning models due to the need for labor-intensive human intervention, large data sets, and resource-intensive z-stacking of raw images.
Innovation Solution
The use of Fourier ptychographic digital refocusing methods to generate uniformly focused images, which are then used to train deep learning models, allowing for automated identification and enumeration of abnormalities in pathology specimens.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If conventional microscopy methods are used to acquire pathology images, then the imaging process is simple, but the focus quality varies across the image and labor-intensive manual refocusing is required
Solution Approach 1:
The system performs preliminary computational refocusing by capturing multiple images at different focal depths and processing them through Fourier ptychographic algorithms before the images are used for deep learning training. This preliminary action creates a library of uniformly focused images that eliminates the need for manual refocusing during analysis, thereby improving both focus quality and productivity
Solution Approach 2:
The patent replaces manual mechanical refocusing operations with an automated computational system based on Fourier ptychographic microscopy. The mechanical adjustment of focus by operators is substituted with algorithmic processing that automatically generates uniformly focused images from multiple captured frames, significantly improving imaging efficiency while maintaining high focus quality
2Measurement precision
If z-stacking of raw images is performed to improve focus coverage, then more focal planes are captured, but data storage requirements and processing resources increase significantly
Solution Approach 1:
The system extracts only the essential focal information needed for uniform focus by processing multiple z-stack images through Fourier ptychographic algorithms. Instead of storing all raw images from multiple focal planes, the method extracts and synthesizes a single uniformly focused image that contains the necessary diagnostic information, thereby reducing data storage requirements while maintaining comprehensive focus coverage
Solution Approach 2:
The computational refocusing method creates a synthetic copy of the specimen at a uniform focal plane by combining information from multiple z-stack images. This synthesized uniformly focused image serves as a representative copy that eliminates the need to store and process all original multi-focal-plane data, reducing storage requirements while preserving diagnostic quality
3Productivity
If deep learning models are trained with non-uniformly focused images, then training data acquisition is faster, but model accuracy and reliability decrease
Solution Approach 1:
The system performs preliminary computational refocusing to generate uniformly focused training images before they are used to train deep learning models. By pre-processing the images to ensure uniform focus quality, the method enables faster acquisition of high-quality training data without compromising model accuracy, as the images are ready for immediate use in training pipelines
Solution Approach 2:
The patent changes the focus parameter of training images from non-uniform to uniform through computational refocusing. By systematically adjusting and standardizing the focus parameter across all training images, the method improves model reliability while maintaining efficient data acquisition, as the automated processing allows rapid generation of uniformly focused images
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
This approach enhances the accuracy and efficiency of deep learning in pathology by providing high-quality, uniformly focused training images, reducing the need for human intervention and minimizing data storage requirements, thereby improving the robustness and predictive accuracy of machine learning models.
Implementation Method 1
Fourier ptychographic digital refocusing procedure
Implementation Method 2
Fourier ptychographic digital refocusing procedure
Data Source
AI summary
Computational refocusing-assisted deep learning methods, apparatus, and systems are described. In certain pathology examples, a representative image is generated using a machine learning model trained with uniformly focused training images generated by a Fourier ptychographic digital refocusing procedure and abnormalities are automatedly identified and/or enumerated based on the representative image.