Heart MRI image multi-task segmentation method based on multi-modal complementary information exploration

Abstract

The invention requests to protect a heart MRI image multi-task segmentation method based on multi-modal complementary information exploration, and the method comprises the steps: S1, sequentially inputting three sequences bSSFP, LGE and T2 of a heart MRI image into a convolutional neural network based on a coding and decoding structure, and extracting the feature information of different sequences in an encoder sharing a weight; S2, recovering the size of a feature map by using a channel reconstruction up-sampling method during decoding, aggregating the features of the three extracted sequences, and sending the features to a corresponding decoder layer through jump connection for feature fusion; and S3, obtaining a classification feature graph through 1 * 1 convolution, and activating the classification feature graph by using a sigmoid function to obtain a final prediction probability graph. The cardiomyopathy can be accurately predicted in combination with the heart multi-sequence image, and the method has certain clinical application value.

Description

technical field [0001] The invention belongs to the technical field of image processing, in particular to a multi-task segmentation method for cardiac MRI images based on multi-modal complementary information exploration. Background technique [0002] Accurate segmentation of myocardial pathology is important for the assessment of myocardial infarction (MI). Cardiac magnetic resonance imaging (MRI) is commonly used in the diagnosis and treatment of cardiovascular diseases, such as myocardial infarction (MI), especially the balanced steady-state free precession (bSSFP) sequence has a clear border of cardiac structures, and late gadolinium-enhanced sequences can enhance representation of infarcted myocardium, while T2-weighted sequences can reveal areas of acute injury and ischemia. Despite great advances in medical imaging techniques, most myocardial pathology segmentation tasks are still done manually, which is a tedious, time-consuming, and error-prone task. Furthermore, ...

AI Technical Summary

Problems solved by technology

[0004] Directly fusing different modal images and inputting them into the network for training may affect the expressive ability of the network due to the difference in intensity distribution between modalities, and designing separate encoders or even decoders for different modalities is likely to cause expensive Computing costs

Images