Pseudo Polarization OCT Image Generation via Machine Learning
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Non-polarization OCT devices cannot acquire polarization OCT images, and existing polarization OCT devices are expensive and complex, limiting their widespread use in medical settings.
Innovation Solution
An information processing device and program that uses machine learning to generate pseudo polarization OCT images from non-polarization OCT images, employing a convolutional neural network to simulate polarization information from input images without polarization data.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Loss of information
If a polarization OCT device is used to acquire polarization OCT images, then polarization information can be obtained, but the device becomes expensive and complex
Solution Approach 1:
The patent applies the copying principle by using a machine learning model to generate pseudo polarization OCT images that replicate the appearance and diagnostic value of true polarization OCT images. Instead of requiring actual polarization-sensitive hardware, the system creates synthetic copies of polarization images from non-polarization input images, thereby obtaining polarization information without the associated hardware complexity
Solution Approach 2:
The patent replaces the mechanical/optical polarization-sensitive detection system with an information processing system based on machine learning. The physical polarization detection mechanism is substituted by an algorithmic approach where a neural network model processes non-polarization images to generate pseudo polarization images, eliminating the need for complex polarization optics while preserving diagnostic capability
2Loss of information
If a polarization OCT device is used to acquire polarization OCT images, then polarization information can be obtained, but the device cost increases
Solution Approach 1:
The patent creates synthetic polarization OCT images through machine learning, generating cost-effective copies that mimic the diagnostic value of expensive true polarization images. This allows medical facilities to obtain polarization information without investing in costly polarization-sensitive OCT hardware
Solution Approach 2:
The patent employs a computationally-generated solution that is significantly cheaper than hardware-based polarization OCT. The machine learning model serves as a low-cost alternative, replacing expensive optical components with software-based image processing that can be deployed on standard computing platforms
3Device complexity
If a non-polarization OCT device is used, then the device remains simple and affordable, but polarization information cannot be acquired
Solution Approach 1:
The patent introduces a machine learning model as an intermediary between non-polarization OCT images and polarization information. This intermediary processing layer extracts and synthesizes polarization-related features from non-polarization input, enabling polarization information acquisition without modifying the original simple and affordable non-polarization OCT hardware
Solution Approach 2:
The patent substitutes the need for mechanical polarization detection components with an information processing system. By using a neural network to generate pseudo polarization images from non-polarization inputs, the system replaces complex polarization optics with software-based processing, maintaining hardware simplicity while recovering polarization information
Data Source
AI summary
An information processing device according to an embodiment includes a learning unit that performs learning on a machine learning model having one or more non-polarization OCT images that are OCT images without polarization information as inputs and a pseudo polarization OCT image corresponding to a polarization OCT image that is an OCT image with polarization information as output; and a storage unit that stores a learning result of the learning unit.