Cross-Device Pathological Data Generation via Annotation Transcription
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Conventional learning models generated using different imaging devices for learning and evaluation fail to accurately evaluate information due to differences in imaging devices, such as resolution and magnification, leading to inadequate diagnosis support.
Innovation Solution
A generation device that transcribes and corrects annotations from a first pathological image captured by one imaging device to a second pathological image using coordinate conversion and adsorption fitting, enabling the generation of a learning model suitable for evaluating images captured by the second device.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Adaptability or versatility
If a learning model is generated using an imaging device for learning that is different from the imaging device for evaluation, then the model can be trained on available data, but the model fails to properly evaluate information captured by the imaging device for evaluation due to differences in imaging characteristics
Solution Approach 1:
The patent transcribes annotations from the first pathological image to the second pathological image, creating a copied version of the annotation data that is adapted to the target imaging device's characteristics. This copying process with transformation allows the learning model to be applied across different devices while maintaining evaluation accuracy.
Solution Approach 2:
The patent transforms annotation parameters (coordinates, dimensions, positions) when transcribing annotations between images captured by different imaging devices. By changing these parameters according to the specific imaging characteristics of each device, the system resolves the contradiction between model versatility and evaluation precision.
2Ease of manufacture
If annotations are directly transferred between images from different imaging devices, then the process is simple, but the transcription accuracy deteriorates due to differences in resolution and magnification
Solution Approach 1:
The patent automatically transforms annotation parameters including coordinates, dimensions, and positions when transcribing between images from different imaging devices. This parameter transformation maintains transcription accuracy while keeping the process automated and simple, resolving the contradiction between ease of manufacture and manufacturing precision.
3Measurement precision
If a learning model is trained on images from a specific imaging device, then the model achieves high accuracy on that device, but the model cannot be effectively applied to images from different imaging devices
Solution Approach 1:
The patent creates a universal annotation transcription system that can adapt annotations and learning models across multiple different imaging devices. By implementing automatic parameter transformation, the system achieves multi-functionality, allowing a single learning model to be effectively applied across different devices while maintaining accuracy.
Solution Approach 2:
The patent transforms annotation parameters to accommodate different imaging device characteristics, enabling the learning model to maintain its high accuracy performance across multiple devices. This parameter adaptation resolves the contradiction between achieving high accuracy on a specific device and maintaining versatility across devices.
Data Source
AI summary
The generation device 100 according to the present application includes an acquisition unit 131 and a generation unit 134. The acquisition unit 131 acquires a first pathological image captured and an annotation that is information added to the first pathological image and is meta information related to the first pathological image. The generation unit 134 generates learning data for evaluating pathological-related information based on a second pathological image according to the second pathological image different from the first pathological image, the learning data being learning data obtained by transcribing the annotation in a manner corresponding to the second pathological image.