Deep Learning 3D Embryo Model from Ultrasound
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current fetal ultrasound images are typically 2D and difficult to convert into 3D stereoscopic images, especially for visualizing fetal faces, due to unclear features like eyes, nose, and mouth, making it challenging to provide realistic 3D fetal models for virtual reality displays.
Innovation Solution
A deep learning-based system that extracts facial features from ultrasound images, generates 3D fetal models, and provides virtual reality content by analyzing fetal images, using a mobile terminal and image providing server to process and synthesize 3D models reflecting per-facial portion features.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Manufacturing precision
If conventional 2D ultrasound image processing methods are used, then the imaging process is simple, but the visualization quality and depth information are insufficient
Solution Approach 1:
The patent applies dimensionality change by converting 2D ultrasound images into 3D fetal face models. The deep learning model processes 2D ultrasound images and generates 3D spatial representations of the fetal face, adding depth information and volumetric visualization capabilities. This transforms the imaging from flat 2D representations to immersive 3D models that can be viewed from multiple angles and provide realistic depth perception.
2Manufacturing precision
If deep learning-based 3D modeling is implemented, then realistic 3D fetal models are generated, but the system complexity and processing requirements increase
Solution Approach 1:
The patent applies segmentation by dividing the fetal face into distinct anatomical regions including eyes, nose, mouth, cheeks, and forehead. The deep learning model processes these segmented facial features independently, extracting specific characteristics from each region. This segmentation approach enables more realistic 3D reconstruction by allowing individual optimization of each facial feature while maintaining overall facial harmony and anatomical accuracy.
3Loss of information
If 3D stereoscopic images are generated from 2D ultrasound images, then depth information is added, but the conversion accuracy is difficult to achieve
Solution Approach 1:
The patent introduces a deep learning model as an intermediary between 2D ultrasound images and 3D fetal face models. This intermediary system learns the complex mapping relationships from training data, enabling accurate inference of depth information and three-dimensional geometry from two-dimensional input. The deep learning model acts as a intelligent translator that bridges the gap between 2D imaging and 3D reconstruction, maintaining high conversion accuracy.
4Ease of operation
If virtual reality display is used for fetal imaging, then user experience is enhanced, but the technical implementation difficulty increases
Solution Approach 1:
The patent creates a virtual copy of the fetal face in three-dimensional space that can be displayed through virtual reality headsets. Instead of requiring complex direct 3D imaging hardware, the system generates photorealistic 3D models that replicate the appearance and features of the actual fetus. This virtual copying approach enables VR display with standard equipment, reducing technical implementation barriers while providing immersive user experience.
Data Source
AI summary
The present invention discloses a virtual reality embryo image providing system. More specifically, the present invention relates to a deep learning-based embryo image providing system which extracts facial features of an embryo from an ultrasound image on the basis of a deep learning technique, generates a 3D model corresponding to the ultrasound image reflecting the facial features, and provides a virtual reality image using the 3D model. According to an embodiment of the present invention, the figure of an embryo can be displayed three-dimensionally through an HMD or the like by setting a plurality of codewords reflecting the features of each body part for a 2D embryo image, and performing a learning procedure on the basis of the codewords according to a deep learning model to provide a 3D model generated by combining the body components that are most similar to the actual face of the embryo. Therefore, a differentiated and realistic embryo imaging service can be provided to a pregnant person.