3D Tissue Classification via Multi-View Image Processing
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current image processing techniques for classifying tissues or lesions in sectional images are inadequate as they often perform discontinuous classifications in the depth direction, leading to inappropriate anatomical classification.
Innovation Solution
An image processing apparatus and method that acquires sectional images in multiple directions, performs primary classification using a discriminator generated through machine learning, and undergoes a secondary classification process to re-specify tissue or lesion types, with corrections based on anatomical features and image feature values to ensure accurate classification across depth directions.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Ease of manufacture
If classification is performed on two-dimensional sectional images, then the classification process is simple, but the classification results are discontinuous in the depth direction and anatomically inappropriate
Solution Approach 1:
The patent transitions from two-dimensional sectional image classification to three-dimensional volumetric data classification. By acquiring multiple sectional images in different directions (axial, coronal, sagittal) and integrating them into 3D space, the system achieves continuous anatomical classification while maintaining computational feasibility through systematic processing of multiple views.
Solution Approach 2:
The patent combines multiple classification results from different sectional directions (axial, coronal, sagittal) into a unified three-dimensional classification. By merging the information from these different perspectives and resolving conflicts through priority rules, the system achieves comprehensive and anatomically accurate tissue/lesion classification.
2Manufacturing precision
If multiple sectional images in different directions are acquired and classified, then classification accuracy improves, but the processing complexity increases
Solution Approach 1:
The patent divides the complex three-dimensional classification task into separate processing stages for different sectional directions (axial, coronal, sagittal). Each direction is processed independently through primary and secondary classification, with results then integrated. This segmentation makes the overall complex process more manageable and systematic.
Solution Approach 2:
The patent performs primary classification for each sectional direction before integrating results. By pre-classifying each view separately and identifying common pixels across multiple views, the system prepares data in advance for the final three-dimensional integration, reducing the complexity of the overall process.
3Productivity
If primary classification is performed on each sectional image independently, then processing is efficient, but classification accuracy decreases due to lack of depth information
Solution Approach 1:
The patent implements a feedback mechanism where primary classification results from multiple sectional directions are evaluated and used to inform secondary classification. Common pixels identified across multiple views provide feedback that refines the classification, improving accuracy while maintaining the efficiency of independent primary processing.
Solution Approach 2:
The patent introduces an intermediary evaluation process that bridges independent primary classifications. By identifying common pixels across multiple sectional images and using them as intermediaries to resolve classifications, the system maintains processing efficiency while achieving accurate three-dimensional classification results.
Data Source
AI summary
The image processing apparatus includes: a sectional image acquisition unit that acquires a plurality of sectional images in a plurality of sectional directions of a test subject; a primary classification unit that performs a primary classification process of specifying, with respect to each of the plurality of sectional images, the type of a tissue or a lesion to which each pixel of each sectional image belongs; a secondary classification unit that performs a secondary classification process of evaluating results of the primary classification process with respect to each of the plurality of sectional images, with respect to a pixel that is common to the plurality of sectional images, to re-specify the type of a tissue or a lesion to which the pixel that is common to the plurality of sectional images belongs; and a correction unit that corrects a result of the secondary classification process on the basis of an anatomic feature of the tissue or the lesion, or an image feature value thereof.