3D Vertebral Body Recognition Using Coronal Projection Numbering
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Conventional deep learning methods for vertebral body identification in medical images face challenges in accurately distinguishing thoracic and lumbar vertebrae due to the presence of ribs in the sagittal plane, particularly affecting accuracy based on age and patient physical conditions.
Innovation Solution
A method and apparatus utilizing a three-dimensional medical image reconstruction based on coronal projection information, employing pre-trained deep learning models to preprocess and project medical images, allowing for precise identification and numbering of vertebral bodies, including lumbar and thoracic areas, through noise reduction, slice thickness normalization, and area differentiation.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Extent of automation
If vertebral body identification is performed based on sagittal plane deep learning models, then the identification process can be automated, but the identification accuracy deteriorates due to rib interference and patient variability
Solution Approach 1:
The patent transitions from two-dimensional sagittal plane analysis to three-dimensional volumetric analysis by reconstructing the medical images into 3D space. This dimensional change allows the system to view vertebral bodies from multiple angles and distinguish them from ribs by analyzing their spatial relationships and morphological characteristics in three dimensions, thereby maintaining automation while significantly improving identification accuracy
2Speed
If conventional deep learning models are used for vertebral body identification, then the processing speed can be maintained, but the reliability deteriorates due to errors from rib interference and patient physical conditions
Solution Approach 1:
By reconstructing medical images into three-dimensional volumetric data, the system gains additional spatial information that enables more reliable differentiation between vertebral bodies and ribs. The 3D representation captures the anatomical relationships and morphological features that are lost in 2D projections, allowing the deep learning model to make more reliable identification decisions while maintaining processing efficiency through optimized volumetric analysis
Solution Approach 2:
The patent modifies the input parameters for the deep learning model by transforming 2D sagittal plane images into 3D volumetric representations with enhanced spatial parameters. This parameter change includes adding depth information, adjusting voxel dimensions, and redefining spatial coordinates, which enables the model to better distinguish vertebral bodies from ribs and account for variations in patient anatomy, thereby improving reliability without sacrificing processing speed
Data Source
Figure 1
Figure 2
Figure 3
AI summary
The present disclosure provides an apparatus of identifying a vertebral body from a medical image, and the apparatus includes a vertebral bone identification module configured to identify the vertebral body based on a multi-slice medical image provided from an outside, in which the vertebral identification module reconstructs the multi-slice medical image to create a three-dimensional medical image, obtains a coronal projection image for the three-dimensional medical image by projecting the three-dimensional medical image in a coronal plane direction, divides the coronal projection image into a selection area including at least one of a lumbar and a thoracic, obtains area information corresponding to the selection area in the three-dimensional medical image based on the divided selection area, and performs numbering on the vertebral body based on the area information and the three-dimensional medical image.