GAN-Based Scanner Harmonization for Digital Pathology Images
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Digital pathology image analysis algorithms are often scanner-specific, leading to compatibility issues when applied to images from different types of scanners, requiring costly and time-consuming updates or modifications, especially when new scanners are introduced.
Innovation Solution
Utilizing Generative Adversarial Networks (GANs) to transform images from one scanner type to have characteristics similar to those of another, allowing existing algorithms to be applied without re-development.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If image analysis algorithms are trained on images from specific whole slide scanners, then the algorithms can achieve high accuracy on those scanner's images, but the algorithms become incompatible with images from different scanner types
Solution Approach 1:
The patent introduces a domain adaptation layer as an intermediary component between the pre-trained image analysis algorithm and images from different scanner types. This layer learns to transform features from source domain (original scanner) to target domain (new scanner), enabling the algorithm to handle multiple scanner types without retraining the entire system. The domain adaptation layer acts as a mediator that bridges the gap between different scanner characteristics.
Solution Approach 2:
The patent changes the parameters of the image analysis algorithm by introducing learnable transformation parameters in the domain adaptation layer. These parameters are optimized during training to adjust the algorithm's behavior for different scanner types. By modifying parameters rather than the entire algorithm structure, the system maintains high accuracy while gaining scanner compatibility.
2Adaptability or versatility
If image analysis algorithms are updated or modified to work with new scanners, then compatibility with new scanners is achieved, but development time and costs increase
Solution Approach 1:
The patent performs preliminary action by pre-training the image analysis algorithm on a source domain scanner and then using domain adaptation to quickly adapt to target domain scanners. Instead of starting from scratch for each new scanner, the system leverages pre-trained knowledge and only trains the lightweight domain adaptation layer, significantly reducing development time for new scanner integration.
Solution Approach 2:
The patent segments the image analysis system into two independent parts: the pre-trained main algorithm and the domain adaptation layer. This segmentation allows the main algorithm to remain fixed and highly accurate, while only the small domain adaptation layer needs to be trained for each new scanner type, reducing overall development time and computational resources required.
3Adaptability or versatility
If image analysis algorithms are updated or modified to work with new scanners, then compatibility with new scanners is achieved, but development costs increase
Solution Approach 1:
The patent segments the system into a fixed pre-trained algorithm and a small domain adaptation layer. This segmentation reduces development costs because the expensive main algorithm is trained only once on the source domain, while the inexpensive domain adaptation layer handles scanner-specific adjustments. This divides the computational and financial burden favorably.
Solution Approach 2:
The patent uses copying by initializing the domain adaptation layer with weights copied from the pre-trained algorithm and then fine-tuning only this small portion. Instead of copying and retraining the entire expensive algorithm for each scanner, the system copies only the necessary adaptation parameters, significantly reducing development costs while maintaining compatibility.
Data Source
AI summary
The present disclosure relates to techniques for transforming digital pathology images obtained by different slide scanners into a common format for image analysis. Particularly, aspects of the present disclosure are directed to obtaining a source image of a biological specimen, the source image is generated from a first type of scanner, inputting into a generator model a randomly generated noise vector and a latent feature vector from the source image as input data, generating, by the generator model, a new image based on the input data, inputting into a discriminator model the new image, generating, by the discriminator model, a probability for the new image being authentic or fake, determining whether the new image is authentic or fake based on the generated probability, and outputting the new image when the image is authentic.