Medical Image Sequence Segmentation for True Positive Pixel Refinement
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Deep learning-based techniques face challenges in accurately segmenting fast-moving and overlapping objects in medical image sequences, such as blood vessels and surgical guide wires, due to the difficulty in distinguishing true positive pixels from false positives.
Innovation Solution
A two-stage neural network approach is employed, where a first ANN focuses on local features for initial segmentation and a second ANN uses contrastive learning to refine segmentations by emphasizing global features and spatial-temporal correspondence, minimizing differences between true positive pixels and maximizing differences between false positives.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If a single neural network is used for medical image segmentation, then the device complexity is low, but the measurement precision of segmentation is insufficient for fast-moving and overlapping objects
Solution Approach 1:
The patent divides the segmentation task into two distinct stages performed by separate neural networks: a first neural network performs initial segmentation to identify candidate pixels, and a second neural network performs refinement to distinguish true positives from false positives. This segmentation of the processing task resolves the contradiction by improving measurement precision through specialized multi-stage processing while managing device complexity through modular architecture.
Solution Approach 2:
The first neural network acts as an intermediary between the raw medical image and the final refined segmentation. It produces an initial segmentation that serves as input to the second neural network, which then refines the results. This intermediary processing stage enables the system to achieve high measurement precision by breaking down the complex segmentation task into manageable steps.
2Measurement precision
If contrastive learning is used to refine segmentation, then the measurement precision improves, but the loss of time increases due to additional processing stages
Solution Approach 1:
The first neural network performs preliminary segmentation by identifying candidate pixels that may belong to the target structure before the second neural network refines the classification. This preliminary action reduces the search space for the second network, allowing it to focus computational resources on distinguishing true positives from false positives among the candidates, thereby improving precision while managing processing time.
Solution Approach 2:
The system performs partial action by having the first neural network identify only candidate pixels rather than classifying all pixels immediately. This partial classification approach allows the second neural network to concentrate computational effort on refining the candidate set, achieving high measurement precision for the critical decision points without unnecessarily processing every pixel at full detail.
3Manufacturing precision
If multiple neural networks are employed for segmentation refinement, then the manufacturing precision of segmentation boundaries improves, but the device complexity increases
Solution Approach 1:
The patent segments the neural network architecture into two specialized components: a first neural network for initial candidate identification and a second neural network for boundary refinement using contrastive learning. This segmentation of functionality improves manufacturing precision by assigning specific tasks to specialized network components, while the modular design manages device complexity through clear separation of concerns and independent training of each network.
Solution Approach 2:
Each neural network is optimized for its specific function: the first network focuses on sensitivity to capture all potential target pixels, while the second network focuses on specificity to refine boundaries and eliminate false positives. This local optimization of network quality for specific tasks improves overall segmentation boundary accuracy while avoiding the need for a single overly complex network.
Data Source
AI summary
An apparatus may obtain a sequence of medical images of a target structure and determine, using a first ANN, a first segmentation and a second segmentation of the target structure based on a first medical image and a second medical image, respectively. The first segmentation may indicate a first plurality of pixels that may belong to the target structure. The second segmentation may indicate a second plurality of pixels that may belong to the target structure. The apparatus may identify, using a second ANN, a first subset of true positive pixels among the first plurality of pixels that may belong to the target structure, and a second subset of true positive pixels among the second plurality of pixels that may belong to the target structure. The apparatus may determine a first refined segmentation and a second refined segmentation of the target structure based on the true positive pixels.