Auto-Encoder DICOM Cube Compression With Embedded Image Encryption
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing medical image compression methods, such as MPEG-type compressions, are inefficient for volumetric medical images due to subtle anatomical changes and lack of motion, leading to potential loss of clinically valuable information and suboptimal storage and transmission processes.
Innovation Solution
An auto-encoder neural network is trained using DICOM cubes to generate encryption and decryption keys, compressing and encrypting medical images by minimizing inter-image entropy and using randomization techniques to ensure robust encryption, allowing secure and efficient storage and transmission.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Loss of substance
If video compressors based on motion prediction are applied to medical images, then compression is achieved, but subtle anatomical changes are lost and clinically valuable information is degraded
Solution Approach 1:
The patent changes the fundamental approach from motion prediction to entropy-based compression. By using auto-encoder neural networks with entropy minimization as the loss function, the system adapts to the specific characteristics of medical images (subtle anatomical changes) rather than applying generic video compression parameters designed for motion-heavy content.
Solution Approach 2:
The patent replaces traditional mechanical compression algorithms (MPEG-type motion prediction) with a learning-based system using auto-encoder neural networks. This substitution allows the system to automatically learn optimal compression strategies for medical images without relying on hand-crafted motion prediction models that are inadequate for subtle anatomical variations.
2Quantity of substance
If standard video compression is used for medical images, then storage and transmission costs are reduced, but clinically valuable information is lost
Solution Approach 1:
The auto-encoder neural network is trained specifically on medical images to learn their unique characteristics. The encoder and decoder components are optimized through self-training to preserve clinically valuable information while achieving compression, making the system self-adapted to medical image requirements rather than relying on generic compression standards.
Solution Approach 2:
The patent implements feedback through the loss function that minimizes inter-image entropy. This feedback mechanism continuously adjusts the compression parameters during training to ensure that subtle anatomical changes are preserved, creating a closed-loop system that optimizes for both compression and information retention.
3Reliability
If encryption is applied to medical images for security, then data security is improved, but storage and transmission efficiency decreases
Solution Approach 1:
The patent merges compression and encryption into a single unified process. The auto-encoder neural network performs both compression (reducing data size) and encryption (securing data) simultaneously, eliminating the need for separate encryption steps that would otherwise increase computational overhead and reduce efficiency.
Solution Approach 2:
The encoder component of the auto-encoder serves multiple functions: it compresses the medical images and simultaneously encrypts them. This multi-functional design allows the system to achieve both security and efficiency goals without requiring separate dedicated encryption hardware or software layers.
Data Source
Figure 1
Figure 2A~2B
Figure 3
AI summary
There is provided an encryption key generation system and method for providing an encryption key and a decryption key to encode and encrypt an input set of medical images, embedded in a DICOM cube. Furthermore, a system and method for compressing and encrypting an input DICOM cube, and a system and method for decompressing and decrypting an input compressed and encrypted DICOM cube are also provided.