Obstacle Data Filtering for Capsule Endoscopy ML
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current machine learning algorithms for medical images, particularly capsule endoscopy, face performance deterioration and increased learning time due to noise images and an unbalanced data set caused by excessive negative data, with no effective method for selectively filtering noise based on the learning objective.
Innovation Solution
A system and method for filtering obstacle data in medical images using a system comprising an obstacle data definition unit, a filter generation unit, and a filtering unit, which defines and generates filters to selectively remove obstacle data based on the machine learning purpose, employing either image processing-based or learning-based filters to classify and remove absolute and relative obstacle data.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Quantity of substance
If machine learning is performed on capsule endoscopic images without filtering noise images, then all available data is used for learning, but learning performance deteriorates and learning time increases
Solution Approach 1:
The patent extracts and removes obstacle data (noise images) from the capsule endoscopic image dataset before performing machine learning. The obstacle data filtering unit identifies and excludes images containing obstacles such as food residues, bubbles, and artifacts that hinder accurate lesion detection, thereby improving learning performance while maintaining an adequate volume of clean training data.
2Quantity of substance
If machine learning is performed on capsule endoscopic images without filtering noise images, then all available data is used for learning, but learning time increases
Solution Approach 1:
The patent extracts and removes obstacle data (noise images) from the capsule endoscopic image dataset before performing machine learning. The obstacle data filtering unit identifies and excludes images containing obstacles such as food residues, bubbles, and artifacts that hinder accurate lesion detection, thereby improving learning performance while maintaining an adequate volume of clean training data.
Solution Approach 2:
The patent performs preliminary filtering of obstacle data from capsule endoscopic images before the machine learning process begins. By pre-processing the dataset to remove noise images and obstacles, the system prepares a cleaner training dataset in advance, which reduces the computational burden during actual learning and decreases overall learning time.
3Quantity of substance
If all capsule endoscopic images are used for learning, then data quantity is maximized, but the data set becomes unbalanced with excessive negative data
Solution Approach 1:
The patent extracts and removes obstacle data (noise images) from the capsule endoscopic image dataset before performing machine learning. The obstacle data filtering unit identifies and excludes images containing obstacles such as food residues, bubbles, and artifacts that hinder accurate lesion detection, thereby improving learning performance while maintaining an adequate volume of clean training data.
Data Source
AI summary
The present disclosure relates to a method for filtering selectively obstacle to be an obstacle to machine learning according to a learning purpose and a system thereof. A system for filtering obstacle data in machine learning of medical images may include an obstacle data definition unit configured to receive definitions of obstacle data according to a machine learning purpose; a filter generation unit configured to generate a filter for filtering the obstacle data; and a filtering unit configured to remove obstacle data in machine learning using the generated filter.