Image Processing Device for Precise Tumor Region Extraction
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing medical image processing methods rely on machine learning results from specific sample images, limiting their ability to accurately extract tumor or lesion regions with varying density values, shapes, and sizes without corresponding sample image data.
Innovation Solution
An image processing device and method that sets a long diameter in a medical image, defines regions around it, calculates evaluation values from density histograms, and uses graph cut techniques to extract specific regions with high precision, independent of specific machine learning results.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If machine learning results from specific sample images are used to extract specific regions, then extraction precision for those specific regions is improved, but the method cannot accurately extract tumor or lesion regions with varying density values, shapes, and sizes without corresponding sample image data
Solution Approach 1:
The system performs self-learning by automatically acquiring multiple candidate specific regions through graph cut processing with different initial values, calculating evaluation values for each, and selecting the optimal region without requiring external sample image data. This self-service mechanism enables the system to adapt to various tumor types independently.
Solution Approach 2:
The system changes the initial value parameter multiple times to generate different candidate specific regions. By varying the initial value and re-executing the graph cut processing, the system explores multiple possible regions and selects the one with the highest evaluation value, thereby adapting to different tumor characteristics without sample data.
2Measurement precision
If multiple candidate specific regions are acquired by changing initial values and performing graph cut processing, then the ability to find the optimal specific region is improved, but the processing time and computational load increase
Solution Approach 1:
The system performs graph cut processing a predetermined number of times with different initial values, which is more than a single processing run but limited to a reasonable count. This partial repetition balances the need for accuracy with acceptable processing time, avoiding excessive computation while ensuring optimal region extraction.
Solution Approach 2:
The system calculates an evaluation value for each candidate specific region obtained through graph cut processing and uses this feedback to select the optimal region. The evaluation value serves as a criterion to determine which candidate best represents the true specific region, enabling efficient selection without exhaustive processing.
Data Source
Figure 1~2
Figure 3~4B
Figure 4C~4D
AI summary
A specific region is extracted from an input image with high precision. Based on a long diameter (L) of a set specific region (RA), a first region (R1) estimated as highly probable to be the specific region (RA) and a second region (R2) estimated as highly probable to be a background region (RB) are set within the input image. Based on a density histogram of each pixel in the first region (R1) and a density histogram of each pixel in the second region (R2), a first evaluation value which indicates a likelihood that a density value represents the specific region (RA) is calculated for each density value. For each pair of two adjacent pixels within an input image (I), based on the first evaluation values respectively corresponding to the two adjacent pixels, a second evaluation value which indicates a likelihood that the two adjacent pixels represent a contour is calculated. Based on the second evaluation value, a connection cost between a plurality of nodes respectively representing each pair of the two adjacent pixels within the input image (I) is set. The plurality of nodes are connected to each other based on the connection cost, to thereby extract the specific region (RA) from the input image (I).