Medical image segmentation task lightweight system construction method

Abstract

The invention proposes a medical image segmentation task lightweight system construction method, and the method comprises the following steps: 1, carrying out the lightweight construction of a network structure, constructing a lightweight network model through employing a depth separable convolution technology, a width network strategy, and a direct scrolling addition method, and reducing the parameter quantity of the network model; step 2, carrying out lightweight construction on the number of network models, and carrying out image segmentation by adopting a one-stage segmentation method from coarse to fine; 3, training the network model by adopting a coarse-to-fine strategy to obtain a lightweight deep learning network model; and step 4, performing image segmentation by using a coarse-to-fine prediction method matched with the first-stage coarse-to-fine segmentation method. Starting from the two aspects of network model structure lightening and network model number lightening, lightening of the whole segmentation system is achieved. Results of the shallow width network and different stages are directly collected and fused, so that resources are saved, and the calculation speed is increased.

Description

technical field [0001] The invention relates to the technical field of deep learning, in particular to a method for constructing a lightweight system for medical image segmentation tasks. Background technique [0002] Convolutional Neural Networks (CNN) quickly surpassed traditional methods based on hand-crafted features (level sets, threshold segmentation, etc.) due to their powerful feature extraction capabilities. made great progress. In addition, the rapid development of CNNs technology has also attracted more and more people to join this research boom, further promoting the development of CNNs technology. However, in the past two years, the CNNs method seems to have entered a relatively bottleneck period, and it is difficult for researchers to make breakthroughs through the innovation of the network itself. Even in many tasks, researchers tend to use some previous research results. For example, in the medical image segmentation task, many studies are often based on s...

AI Technical Summary

Problems solved by technology

Aiming at the problems existing in the existing methods, the present invention proposes a method for constructing a lightweight system for medical image segmentation tasks to solve related problems in the process of medical image segmentation

Images