Learned Lung Deformation Model for Micronodule Positioning
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing methods for registering pre-surgical dye sites with CT images during surgery are burdensome for doctors and patients, and struggle with accurately simulating patient-specific mechanical properties and deformations, particularly in the lungs which change from aerated to deaerated conditions during surgery.
Innovation Solution
A learned model is generated using machine learning to estimate lung deformation with high accuracy by learning deformation per micro region, such as vertex units, rather than the entire shape, incorporating upsampling to increase data and using kernel regression models to predict lung deformation during surgery.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If the entire shape is used as the learning unit for deformation, then the model can handle complex overall deformation, but the estimation accuracy for local regions decreases
Solution Approach 1:
The patent divides the learning process into two levels: (1) learning the deformation of the entire shape as a whole, and (2) learning the deformation of micro regions (vertices) within the shape. This segmentation allows the model to capture both global deformation patterns and local deformation details, thereby improving estimation accuracy without overwhelming complexity
Solution Approach 2:
The patent applies different learning strategies to different parts of the shape. The entire shape is learned using one approach while micro regions (vertices) are learned using another approach. This local quality principle allows the model to focus computational resources on capturing fine-grained local deformations where precision is most critical, while still maintaining awareness of global shape changes
2Measurement precision
If more measurement data items are collected to improve model accuracy, then the estimation precision increases, but the burden on doctors and patients increases
Solution Approach 1:
The patent uses upsampling to generate additional training data points from existing measurement data. Instead of collecting more physical measurement data (which increases burden), the system creates synthetic training examples by upsampling the available data, thereby improving model accuracy without requiring additional measurements from patients or doctors
Solution Approach 2:
The patent creates synthetic training data by copying and transforming existing measurement data through upsampling techniques. This allows the model to learn from multiple variations of the same underlying data, effectively increasing the training dataset size without requiring additional physical measurements, thus maintaining ease of operation while improving model accuracy
Data Source
Figure 1
Figure 2
Figure 3
AI summary
A machine learning device includes: a generation unit generating a first shape model representing a shape of an object before deformation and a second shape model representing a shape of the object after the deformation based on measurement data before and after the deformation; and a learning unit learning a feature amount including a difference value between each micro region and another micro region that constitute the first shape model, and a relation providing a displacement from the each micro region of the first shape model to each corresponding micro region of the second shape model.