Information processing device, information processing method, and information processing program
The neural network circuit subdivides high-resolution 3D images into smaller, higher-resolution ranges, using size-adjustment processing to enhance recognition accuracy and reduce memory requirements.
JP2026084547APending Publication Date: 2026-05-21株式会社MEDALVISION
View PDF 2 Cites 0 Cited by
Patent Information
- Authority / Receiving Office
- JP · JP
- Patent Type
- Applications
- Current Assignee / Owner
- 株式会社MEDALVISION
- Filing Date
- 2024-11-11
- Publication Date
- 2026-05-21
AI Technical Summary
Technical Problem
High-resolution 3D images require significant memory resources for processing, leading to expensive hardware or hardware that does not exist, and subdividing images for processing results in loss of context recognition.
Method used
A neural network circuit with multiple layers processes high-resolution images by subdividing them into smaller, higher-resolution ranges, using a size-adjustment processing circuit to crop and enlarge outputs between neural network circuits for accurate recognition.
Benefits of technology
Enables accurate inference on entire high-resolution 3D images without requiring excessive memory, maintaining positional relationships and improving recognition accuracy.
✦ Generated by Eureka AI based on patent content.
Smart Images
Figure 2026084547000001_ABST
Abstract
The present invention provides an information processing device capable of inference from high-resolution medical images, including inference from the entire medical image. [Solution] A neural network circuit for semantic segmentation targeting a three-dimensional image having multiple layers, comprising multiple neural network circuits of the same or similar shape, wherein a first neural network circuit among the multiple neural network circuits processes a first range image of the three-dimensional image, and a second neural network circuit among the multiple neural network circuits processes a second range image of the three-dimensional image, which is narrower than the range of the first range image and has a higher resolution than the resolution of the first range image, and a size-adjustment processing circuit is configured to perform a cropping process and an enlargement process on the output of one or more arbitrary blocks in the first neural network circuit to cut out a predetermined range, and to couple it to the output of the corresponding block in the second neural network circuit.
Need to check novelty before this filing date? Find Prior Art