Mass Image Quality Conversion for Low-Signal Noise Reduction
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
The challenge in improving the image quality of mass images is exacerbated by the small amount of ions acquired from each micro area, leading to low signal-to-noise ratios and difficulty in applying machine learning techniques due to the scarcity of high-definition images, resulting in blurred and noisy mass images.
Innovation Solution
A three-stage process involving a pre-processor, image quality converter, and post-processor is employed to enhance mass image quality using a machine learning model. The pre-processor adjusts the mass image to fit the input conditions of the converter, the converter applies an image quality conversion model, and the post-processor adjusts the output to meet mass image specifications, including noise correction and intensity scaling.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Object-affected harmful factors
If a simple smoothing filter is applied to reduce noise in the mass image, then the noise or roughness becomes less prominent, but the entirety of the mass image itself becomes blurred
Solution Approach 1:
The patent introduces a machine learning model as an intermediary between the noisy mass image and the final processed image. This model learns the mapping from noisy to clean images through training, acting as a smart mediator that selectively removes noise while preserving important structural information, unlike traditional filters that uniformly blur the entire image.
Solution Approach 2:
The patent changes the approach from deterministic filtering to probabilistic learning by training a machine learning model on multiple training images. The model learns optimal noise removal parameters through exposure to various noise patterns and image structures, enabling adaptive noise reduction that preserves image quality.
2Measurement precision
If machine learning techniques are applied to improve mass image quality, then image quality can be enhanced, but it is difficult to produce an image quality conversion model because high-definition mass images are scarce
Solution Approach 1:
The patent creates synthetic training data by copying and transforming existing mass images. Training images are generated by adding simulated noise to clean reference images, creating artificial training datasets that mimic real noisy conditions without requiring actual high-definition mass images, thus solving the data scarcity problem.
Solution Approach 2:
The patent uses an intermediary training process where the machine learning model learns from synthetic noisy-clean image pairs. This intermediary training stage prepares the model to handle real noisy mass images, bridging the gap between scarce real data and the need for robust noise removal capability.
3Measurement precision
If the amount of ions acquired from each micro area is increased to improve S/N ratio, then image quality improves, but the acquisition time and measurement conditions become more restrictive
Solution Approach 1:
The patent replaces the mechanical approach of increasing ion acquisition (which requires longer measurement times and stricter conditions) with an information processing approach using machine learning. The model computationally enhances the S/N ratio by learning from training data, substituting physical signal enhancement with intelligent signal processing.
Data Source
AI summary
A pre-processor applies a pre-process to an original mass image produced through mass spectrometry of a sample, to produce a model input image. An image quality converter has an image quality conversion model produced through machine learning based on a group of images produced by a scanning electron microscope, and produces a model output image through image quality conversion of the model input image. A post-processor applies a post-process to the model output image, to produce a mass image after image quality conversion.