Machine Learning Model for Blood Circulation Anomaly Inference
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current methods for diagnosing retinal diseases like diabetic retinopathy require extensive annotation and large datasets for machine learning models, making it time-consuming and inefficient to identify blood circulation anomalies in medical images without performing fluorescein eye-fundus angiography, which also poses risks due to contrast dye.
Innovation Solution
An information processing device and method that uses machine learning models to infer blood circulation anomalies by inputting medical images with annotation information and derived blood vessel images, employing loss functions and neural networks to efficiently learn and infer anomalies with a smaller dataset, avoiding dominance by easy-to-infer areas and incorporating ordinal scales and probability distributions.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If annotation information is provided to eye-fundus images using fluorescein eye-fundus angiography, then blood circulation anomalous areas can be accurately identified, but the workload and time required for data collection increases significantly
Solution Approach 1:
The patent introduces an auxiliary model that generates blood vessel images as an intermediary representation. This auxiliary model processes eye-fundus images to create simplified blood vessel maps, which then serve as additional input features for the main model to identify blood circulation anomalies. This intermediary step reduces the need for time-consuming manual annotation while maintaining diagnostic accuracy.
Solution Approach 2:
The patent performs preliminary processing by generating blood vessel images from eye-fundus images before the main anomaly detection process. This preliminary action of creating auxiliary blood vessel representations prepares the data in advance, reducing the subsequent workload for anomaly identification and decreasing overall processing time.
2Measurement precision
If a large amount of annotated learning data is collected, then the learned model achieves high accuracy in inferring blood circulation anomalies, but the complexity and cost of data annotation increases
Solution Approach 1:
The patent implements a self-service mechanism where the auxiliary model automatically generates blood vessel images from raw eye-fundus images without requiring manual annotation. This self-generated auxiliary data serves as additional training input, reducing the need for extensive manually annotated datasets while maintaining model accuracy.
Solution Approach 2:
The patent creates a simplified copy or representation of the blood vessel structure through the auxiliary model. This copied blood vessel image serves as an auxiliary input that captures essential vascular information without requiring complex manual segmentation, thereby reducing annotation complexity while preserving diagnostic value.
3Measurement precision
If manual annotation of blood circulation anomalies is performed by specialists, then accurate learning data can be obtained, but the workload on medical workers increases
Solution Approach 1:
The auxiliary model acts as an intermediary that automatically generates blood vessel images, reducing the direct burden on medical workers. This intermediary processing step handles the complex task of vascular representation, allowing specialists to focus on higher-level diagnostic interpretation rather than detailed annotation work.
Solution Approach 2:
The patent replaces the mechanical process of manual annotation by specialists with an automated computational system (auxiliary model). This substitution eliminates the need for medical workers to perform time-consuming annotation tasks while maintaining or improving annotation consistency and accuracy.
Data Source
AI summary
An information processing device, an information processing method, and a computer-readable recording medium that are capable of generating, with a smaller amount of learning data, a learned model that infers a blood circulation anomalous area in a medical image are provided.A learning unit 124 and a model output unit 126 are provided. The learning unit 124 is configured to cause a machine learning model 125 to learn by inputting medical images and blood vessel images into the machine learning model, the medical images being provided with annotation information of a blood circulation anomalous area, the blood vessel images being obtained by estimating a blood vessel area in the medical images based on the medical images. The model output unit 126 outputs a learned model having learned at the learning unit.