Endoscope Image Recognition Neural Network Filtering
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current medical endoscope image recognition methods are not robust enough to adapt to the entire process of capturing images, as they are affected by endoscope movement and exposure to liquids and foreign matters, leading to weak classification prediction and interference in image data.
Innovation Solution
A medical endoscope image recognition method and system that uses a neural network to filter out interference, identify organ information, and switch to appropriate imaging modes for accurate lesion region localization and categorization, enhancing robustness and accuracy through deep learning techniques.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Adaptability or versatility
If a single classification prediction function is used for medical image recognition, then the system structure remains simple, but the system cannot adapt to the whole process of photographing by the endoscope and shows weak robustness
Solution Approach 1:
The patent segments the image recognition system into multiple independent modules: quality assessment module, organ part recognition module, imaging type identification module, lesion region localization module, and lesion category identification module. Each module performs a specific function in the endoscope image processing pipeline, allowing the system to handle the complex photographing process through coordinated operation of simpler, specialized components rather than a single complex unified system.
Solution Approach 2:
The patent implements a multi-functional recognition system where a single integrated framework performs multiple tasks including quality assessment, organ part identification, imaging type determination, lesion localization, and lesion classification. This universal system adapts to the whole photographing process by sequentially executing different functions based on the endoscope operation context, achieving versatility without requiring separate dedicated systems for each function.
2Productivity
If endoscope images are captured during normal operation, then the imaging process is continuous, but the images are affected by switching and shaking and encounter liquids and foreign matters resulting in interference and noise
Solution Approach 1:
The patent applies preliminary action by first assessing the quality of captured endoscope images before proceeding to recognition and analysis. The quality assessment module evaluates image parameters such as focus, brightness, and noise levels in advance, allowing the system to identify and handle low-quality images caused by switching, shaking, or foreign matter interference before they enter the main recognition pipeline, thus preventing unreliable data from affecting subsequent processing.
Solution Approach 2:
The patent converts the harmful effects of switching and shaking into beneficial filtering opportunities. Instead of simply rejecting corrupted images, the system uses the quality assessment results to selectively process images, applying correction algorithms or adjusting recognition parameters based on the type and severity of interference detected. This transforms the noise and interference caused by operational challenges into information that improves the robustness of the overall recognition system.
3Measurement precision
If deep learning classification prediction is applied to endoscope images, then classification accuracy improves, but the system becomes vulnerable to interference and noise from endoscope movement and foreign matters
Solution Approach 1:
The patent introduces quality assessment as an intermediary step between image capture and deep learning classification. This intermediary module evaluates image quality metrics and acts as a filter, determining whether the image is suitable for direct classification or requires preprocessing. By placing this intermediary quality check before the classification algorithm, the system protects the accurate but vulnerable deep learning model from processing noisy or degraded images, thus maintaining classification precision while mitigating the impact of movement and foreign matter interference.
Data Source
AI summary
A medical endoscope image recognition method is provided. In the method, endoscope images are received from a medical endoscope. The endoscope images are filtered with a neural network, to obtain target endoscope images. Organ information corresponding to the target endoscope images is recognized via the neural network. An imaging type of the target endoscope images is identified according to the corresponding organ information with a classification network. A lesion region in the target endoscope images is localized according to an organ part indicated by the organ information. A lesion category of the lesion region in an image capture mode of the medical endoscope corresponding to the imaging type is identified.