Medical Image Segmentation via Temporal Dynamic Processing
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current medical image segmentation methods using deep learning networks require extensive labeling of medical images at multiple time points, leading to a heavy workload and increased complexity in model training.
Innovation Solution
A medical image segmentation method that processes a set of medical images in a time dimension to create a temporal dynamic image, which is then used to extract target region features using a medical image segmentation model, reducing the need for extensive labeling.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If deep learning network model is trained by using medical images at all time points, then segmentation accuracy is improved, but labeling workload increases heavily
Solution Approach 1:
The patent segments the training process into two distinct phases: (1) pre-training the deep learning network model using only baseline medical images (reducing initial labeling requirements), and (2) fine-tuning the pre-trained model using medical images from all time points (achieving high segmentation accuracy). This segmentation of training stages resolves the contradiction by distributing labeling requirements across different phases rather than requiring full labeling upfront.
Solution Approach 2:
The patent performs preliminary action by pre-training the deep learning network model on baseline medical images before the actual segmentation task. This pre-training phase establishes a foundational model that can be later fine-tuned with time-point specific images, reducing the immediate labeling burden while maintaining the capability for accurate segmentation when full data becomes available.
2Measurement precision
If deep learning network model is trained by using medical images at all time points, then segmentation accuracy is improved, but model training complexity increases
Solution Approach 1:
The training process is segmented into distinct stages: pre-training on baseline images and fine-tuning on time-point images. This segmentation simplifies the overall training complexity by breaking down a single complex training process into manageable phases, each with specific objectives and data requirements.
Solution Approach 2:
The patent maintains continuity of useful action through the fine-tuning process, where the pre-trained model is continuously improved using time-point specific medical images. This continuous refinement approach ensures that the model adapts to temporal variations while building upon the foundational knowledge acquired during pre-training, achieving high accuracy without restarting the entire training process.
Data Source
AI summary
The present disclosure provides a medical image segmentation method. The medical image segmentation method includes acquiring a to-be-processed medical image set, the to-be-processed medical image set including a plurality of to-be-processed medical images corresponding to different time points, processing the to-be-processed medical image set in a time dimension according to the to-be-processed medical images and the time points corresponding to the to-be-processed medical images to obtain a temporal dynamic image, and extracting a target region feature from the temporal dynamic image by using a medical image segmentation model, to acquire a target region.