Adaptive FOV Image Segmentation for Multi-Scale Targets
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Medical image segmentation faces challenges in achieving precise results due to large differences in scales and shapes of targets, leading to insufficient precision for small targets and incomplete segmentation for large targets, especially in fully-automatic and semi-automatic methods.
Innovation Solution
An image segmentation apparatus and method that calculates an adaptive Field of View (FOV) function based on statistical shape features of targets, generating patches corresponding to these scales, and using a trained model for segmentation, allowing for precise segmentation across multiple scales.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Area of stationary object
If a large Field of View (FOV) is designated for segmentation, then coverage of large targets is improved, but precision levels for small targets become insufficient and computation resources are wasted
Solution Approach 1:
The patent implements dynamic FOV adjustment by training the model to predict optimal FOV sizes based on input image characteristics. The system transitions from static FOV designation to dynamic adaptation, where the FOV is automatically adjusted according to the scale and characteristics of targets in each specific image, resolving the contradiction between coverage and precision.
Solution Approach 2:
The patent changes the FOV parameter dynamically based on image content. By introducing FOV prediction as an additional output of the neural network, the system automatically adjusts the FOV parameter to match the scale of targets, enabling high precision segmentation across varying target sizes without manual intervention.
2Loss of energy
If a small Field of View (FOV) is designated for segmentation, then computation resources are saved, but coverage of large targets becomes incomplete
Solution Approach 1:
The system dynamically determines the appropriate FOV size for each image based on target characteristics. This eliminates the need to use a universally large FOV that wastes computation resources on images containing only small targets, while still ensuring adequate coverage when large targets are present.
Solution Approach 2:
The FOV parameter is changed adaptively based on the scale of targets detected in the image. The model predicts the optimal FOV and adjusts this parameter accordingly, optimizing the balance between computational efficiency and segmentation coverage for each specific case.
3Device complexity
If a fixed FOV is used for all segmentation tasks, then device complexity is reduced, but segmentation accuracy for multi-scale targets deteriorates
Solution Approach 1:
The patent creates a universal segmentation system that handles multi-scale targets through a single model capable of both segmentation and FOV prediction. This multi-functional approach maintains relatively simple device architecture while achieving high accuracy across varying target scales, avoiding the need for multiple specialized models.
Solution Approach 2:
The introduced FOV prediction module acts as an intermediary between the input image and the segmentation process. This mediator analyzes image characteristics and determines the appropriate FOV, enabling the segmentation model to achieve high accuracy for multi-scale targets without requiring complex multi-model architectures.
Data Source
AI summary
An image segmentation apparatus according to an embodiment includes processing circuitry configured: to calculate a variable field-of-view mathematical function capable of adaptively generating fields of view having corresponding sizes, with respect to a plurality of segmentation targets included in an image; to generate patches having corresponding sizes, with respect to the plurality of segmentation targets, by using the variable field-of-view mathematical function; and to obtain a segmentation result of the plurality of segmentation targets, by carrying out an inference on the image while using a segmentation model trained with the patches.