Ophthalmic Image Processing Device Using Machine Learning for Tissue Boundary Detection
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing ophthalmic image processing techniques struggle to accurately detect tissue boundaries and specific parts, especially when tissue structures are collapsed due to disease, and they also fail to quantify the degree of abnormality in tissue structures effectively.
Innovation Solution
An ophthalmic image processing device that uses a mathematical model trained with a machine learning algorithm to acquire a probability distribution of tissue boundaries and specific parts within an ophthalmic image, allowing for accurate detection and quantification of structural abnormalities.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If traditional boundary detection methods are used, then the process is simple, but detection accuracy is reduced when tissue structure is collapsed
Solution Approach 1:
The mathematical model is trained in advance with normal ophthalmic images to learn the mapping between image features and tissue boundary coordinates. This preliminary training enables the model to accurately detect boundaries even when tissue structures are collapsed or degraded, without requiring complex real-time processing adjustments.
Solution Approach 2:
The patent replaces traditional mechanical image processing methods (filtering, edge detection algorithms) with a machine learning-based mathematical model. This substitution allows the system to automatically learn complex patterns and relationships in ophthalmic images, achieving higher detection accuracy for collapsed tissue structures.
2Measurement precision
If GAN-based anomaly detection is used, then abnormal parts can be identified, but the degree of abnormality cannot be quantified
Solution Approach 1:
The mathematical model outputs probability distributions that quantify the likelihood of tissue boundaries at specific locations. By analyzing deviations of these probability distributions from expected values, the system quantifies the degree of abnormality in tissue structures, providing users with numerical metrics rather than just binary anomaly detection.
3Measurement precision
If mapping-based layer detection is used, then layer thickness can be measured, but accuracy is lost when tissue structure collapses
Solution Approach 1:
The mathematical model is pre-trained with a large dataset of normal ophthalmic images, learning the typical patterns and variations of healthy tissue structures. This preliminary action enables the model to recognize deviations from normality and accurately detect boundaries even when tissue structures are collapsed or distorted by disease.
Solution Approach 2:
The system uses probability distribution outputs from the mathematical model to continuously assess detection confidence. When detection reliability is low (e.g., in severely collapsed tissue), the system can identify these cases and potentially request additional imaging or expert review, improving overall system reliability.
Data Source
AI summary
In this invention, a control unit in an ophthalmic image processing device acquires an ophthalmic image captured by an ophthalmic image capture device (S11). The control unit, by inputting the ophthalmic image into a mathematical model that has been trained by a machine-learning algorithm, acquires a probability distribution in which the random variables are the coordinates at which a specific site and/or a specific boundary of a tissue is present within a region of the ophthalmic image (S14). On the basis of the acquired probability distribution, the control unit detects the specific boundary and/or the specific site (S16, S24).