An MRI brain tumor image segmentation method based on DBN neural network

Abstract

The invention discloses an MRI brain tumor image segmentation method based on a DBN neural network. Firstly, a plurality of images are picked out from an existing patient brain MRI sequence image library as training samples, and the training samples are preprocessed and the significance map is calculated. Then the down-sampling is sent to DBN neural network for unsupervised and supervised training. Aiming at the extreme imbalance of training samples, the down-sampling is carried out in the non-tumor area, which improves the detection rate of positive samples. After the training, the test images to be segmented can be sent to the network for segmentation, and visual attention model is introduced to enhance the segmentation accuracy of the network for the more difficult segmentation area, and the segmentation results are output at last.

Description

technical field [0001] The invention belongs to the technical field of image processing, and in particular relates to a method for segmenting MRI brain tumor images based on a DBN neural network. Background technique [0002] In recent years, brain tumors have become one of the tumors with the highest morbidity and mortality. Magnetic resonance imaging (MRI) can perform high spatial resolution and high contrast imaging of brain soft tissue, and is the best choice for doctors to analyze brain structure, so it is widely used clinically. In the brain MRI image processing, the precise segmentation of the tumor part is a crucial step, which plays a vital role in the doctor's subsequent analysis and judgment. At present, this step still relies heavily on manual segmentation, which is very time-consuming and has strong instability. Therefore, it is of high practical value to find an accurate automatic segmentation method. However, due to the highly variable shape, location and st...

AI Technical Summary

Problems solved by technology

However, this type of method often needs to manually extract features in advance, so it requires the designer to have relevant professional knowledge, which is not realistic in many cases, and the manually extracted features have the disadvantages of strong pertinence and poor scalability.

Images