Medical Image Anonymization via Segmentation and Pattern Replacement
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing pseudonymization technologies for medical images lack the ability to efficiently preserve anatomical information, leading to loss of diagnostic information and inability to provide incidental findings.
Innovation Solution
A method and apparatus for anonymizing medical images by separating identification and de-identification areas, generating a de-identified image using pattern areas with randomized signal intensity values, and replacing the de-identification area with the de-identified image, thereby preserving anatomical information.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If simple image processing techniques such as masking or blurring are applied to specific areas in medical images, then personal information is de-identified, but the reality of the pseudonymized image is low and anatomical information is lost
Solution Approach 1:
The patent segments the medical image into multiple regions: identification areas (preserving anatomical structures), de-identification areas (containing personal information), and pattern areas (with randomized signal intensities). This segmentation allows selective processing that de-identifies personal information while preserving anatomical information for diagnostic purposes.
Solution Approach 2:
Different regions of the medical image are assigned different quality characteristics. Identification areas maintain original signal intensities and anatomical details, while de-identification areas are transformed with randomized patterns. This local differentiation ensures that anatomical information is preserved where needed while personal information is effectively de-identified.
2Reliability
If direct editing of finally generated facial images, coronal images, or sagittal images is performed, then de-identification is achieved, but overall diagnostic information for areas masked by de-identification areas is lost
Solution Approach 1:
The patent performs preliminary segmentation to identify and separate de-identification areas from identification areas before applying any editing operations. By pre-defining pattern areas with randomized signal intensities in de-identification regions while preserving identification areas, the method prevents loss of diagnostic information that would occur with direct editing of final images.
3Reliability
If de-identification areas are masked or blurred, then personal information is hidden, but incidental findings based on anatomical information cannot be provided
Solution Approach 1:
The patent applies different quality treatments to different regions: de-identification areas receive randomized pattern transformations to hide personal information, while identification areas retain original anatomical information. This local quality differentiation enables both high anonymization levels and preservation of diagnostic utility for incidental findings.
Solution Approach 2:
By segmenting the image into de-identification areas and identification areas, the patent allows independent processing of each region. The identification areas preserve anatomical information necessary for diagnostic purposes and incidental findings, while de-identification areas are transformed to protect personal information.
Data Source
Figure 1
Figure 2
Figure 3
AI summary
Disclosed are a method of anonymizing medical images, the method comprising: acquiring or receiving a first medical image; separating an identification area containing anatomical structure information and a de-identification area containing a skin area in the first medical image; generating a de-identified image including a plurality of pattern areas with a predetermined size for the de-identification area; and generating a second medical image by replacing the de-identification area with the de-identified image.