3D Volume Projection Labeling for Multi-View AI Training
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
The labor-intensive process of manually labeling large datasets for supervised machine learning, particularly in medical imaging, where many images represent slices of the same three-dimensional volume, hinders efficient training of AI systems.
Innovation Solution
A method and system that define a three-dimensional volume in a common frame of reference for a set of two-dimensional images, projecting this volume onto each image and using image metrics like surface area, standard deviation, and pixel value histograms to automatically label images based on their inclusion within the three-dimensional volume, reducing manual intervention.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If manual labeling is used for each image in the dataset, then labeling accuracy can be ensured, but the time and labor required increases significantly
Solution Approach 1:
The method segments the labeling task by introducing 3D volume definitions that group multiple 2D images together. Instead of labeling each image individually, a single 3D volume label applies to all images within that volume, dividing the overall labeling workload into manageable segments based on spatial relationships.
Solution Approach 2:
The patent transitions from 2D image-level labeling to 3D volume-level labeling by adding a temporal or depth dimension. This dimensional expansion allows multiple 2D slices to be consolidated under a single 3D volume label, reducing the total number of labeling operations required while maintaining accuracy through spatial context.
2Reliability
If more images are manually labeled to improve training data quality, then AI system performance improves, but the labor intensity and cost increase
Solution Approach 1:
A single 3D volume label serves multiple functions by simultaneously labeling all 2D images that fall within that volume. This universal labeling approach ensures consistent annotation across multiple views or time points while eliminating redundant manual labeling efforts, thereby improving productivity without compromising data quality.
Solution Approach 2:
The method performs preliminary action by defining 3D volumes and their spatial boundaries before the actual labeling process. This pre-establishment of spatial context allows the system to automatically determine which images belong to which volumes, reducing the need for repeated manual assessment and improving overall labeling efficiency.
3Measurement precision
If detailed pixel-level labeling is performed to achieve high precision, then the quality of training data improves, but the complexity of the labeling process increases
Solution Approach 1:
The patent extracts the essential spatial boundaries by defining 3D volume extents rather than requiring detailed pixel-level annotations. This extraction approach captures the critical information needed for training while eliminating the complexity of manual pixel-by-pixel labeling, thereby reducing process complexity while maintaining sufficient precision for AI training purposes.
Data Source
AI summary
A method includes receiving a plurality of images, each of the plurality of images being a two-dimensional view of a portion of a three-dimensional object; receiving a definition of a three-dimensional volume in a same frame of reference as the plurality of images; receiving a label for the three-dimensional volume; projecting the three-dimensional volume onto each of the plurality of images; determining, for each of the plurality of images an image metric; and determining whether to assign the label to respective ones of the plurality of images based on the image metric.