Biometric Image Comparison Using Adversarial Noise for Diagnosis
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing machine learning models for reading medical images, such as biometric images, suffer from inaccuracies due to insufficient learning data and environmental differences, leading to incorrect disease predictions.
Innovation Solution
An apparatus and method that uses a Generative Adversarial Network (GAN) to generate a second biometric image with altered feature information by applying adversarial noise to the original image, allowing for accurate comparison and explanation of the reading results.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Extent of automation
If machine learning models are used to extract features from biometric images, then disease prediction capability is improved, but prediction accuracy deteriorates due to insufficient learning data and environmental differences
Solution Approach 1:
The patent introduces an explanation generation module as an intermediary between the machine learning model and the practitioner. This module generates human-understandable explanations for model predictions, addressing the accuracy issue by making the model's decision-making process transparent and verifiable, thereby compensating for insufficient learning data and environmental variations
Solution Approach 2:
The system implements feedback by providing practitioners with explanations of model predictions and allowing them to input correction information when predictions are incorrect. This feedback loop enables continuous improvement of the model's accuracy by learning from practitioner corrections and adjusting future predictions accordingly
2Adaptability or versatility
If standard machine learning models are deployed across different medical environments, then adaptability is improved, but reliability deteriorates due to environmental differences and imaging device variations
Solution Approach 1:
The patent changes the parameter representation by generating environment-specific explanation patterns. The explanation generation module adapts its explanation style and content based on the specific medical environment and imaging device being used, allowing the system to maintain reliable predictions across diverse environments by adjusting how predictions are explained and validated
Solution Approach 2:
The system performs preliminary action by generating explanations and validation information before final disease prediction is made. This allows practitioners to verify predictions against environment-specific patterns and conditions, ensuring reliability before accepting the diagnosis
3Productivity
If feature extraction is performed using machine learning models, then reading efficiency is improved, but explanation clarity deteriorates making it difficult to understand prediction reasons
Solution Approach 1:
The explanation generation module serves as an intermediary that translates the machine learning model's internal feature extraction processes into human-understandable explanations. It bridges the gap between efficient automated feature analysis and clear communicable results, preserving explanation information in a format practitioners can understand and verify
Solution Approach 2:
The patent replaces the traditional mechanical explanation approach (direct model output) with an intelligent explanation generation system that uses natural language processing and knowledge graphs to create comprehensible explanations, maintaining both efficiency and clarity
Data Source
AI summary
Provided are a device, a method, and a system for supporting biometric image finding/diagnosis, the device comprising: a processor; and a memory including one or more instructions implemented to be executed by the processor, wherein the processor; extracts a first attribute information from a first biometric image of an object on the basis of a machine learning model; changes the first attribute information of the first biometric imaged by mapping adversarial noise to the first biometric image, so as to generate a second biometric image having second attribute information; and displays the first biometric image having the first attribute information and the second biometric image having the second attribute information on a display unit.