SP-FCN-based MRI brain tumor image segmentation system and method

Abstract

The invention provides an SP-FCN-based MRI brain tumor image segmentation system and method. The method comprises the steps of extracting features of brain tumor images in different modes through a superpixel method to obtain ROI segmentation results under different modals; taking ROI images under different modals as the input data of the FCN, carrying out data fusion to obtain a final segmentation result, and taking the ROI images under different modals as the input, so that the calculation amount of the data is reduced, and the final result is more accurate; meanwhile, the difference of white matter lesion images can be more easily captured by adopting multi-mode image fusion than a single-mode image, and the stability of the algorithm can be improved by multiple registration.

Description

technical field [0001] The disclosure belongs to the field of medical image processing and deep learning, and in particular relates to a SP-FCN-based MRI brain tumor image segmentation system and method. Background technique [0002] The statements in this section merely provide background information related to the present disclosure and do not necessarily constitute prior art. [0003] Regarding the application of MRI brain tumor image segmentation methods, the development in the medical field is relatively mature. With the improvement of people's awareness of life safety and the wide application of medical image processing technology in the fields of pattern recognition, physics and computer, higher requirements have been put forward in terms of accuracy and speed of brain tumor image segmentation. Therefore, in order to achieve real-time and accurate segmentation of brain tumor images and meet the requirements of brain tumor image segmentation, it is necessary to establ...

AI Technical Summary

Problems solved by technology

[0004] The inventors found that in the prior art, for the MRI brain tumor image segmentation problem, it is mainly realized by using deep learning algorithm, and the general processing flow is to pre-image the image Processing, extracting useful features, and finally training the model based on the extracted features to achieve segmentation; for MRI brain tumor image segmentation, commonly used methods include segmentation methods based on regions, thresholds, and edge detection. Region-based segmentation methods are more sensitive to noise , the threshold-based segmentation method is a difficulty in selecting the optimal threshold, and the segmentation method based on edge detection has low segmentation accuracy. In addition, due to the complexity and changeability of MRI images, in the process of extracting image features, there may be low tissue contrast in the image , Artifact interference, which affects the accuracy of subsequent image segmentation; at the same time, for the MRI brain tumor image segmentation problem, although the traditional convolutional neural network has good generalization ability, the convergence speed in the model training process Too slow, resulting in too long training time. In addition, there are problems such as easy to fall into local minima, slow convergence speed, and strong dependence on initial parameters. Especially for some deep neural networks, there may be gradient explosion or Vanishing gradient etc.

Images