Medical image modal synthesis method

Abstract

The invention discloses a medical image modal synthesis method. Source modal image space resolution normalization processing is carried out, resolutions of images are made to be the same through normalization, source modal image gray value standardization or normalization processing is carried out, normalization of source modal image gray values after first-step processing is carried out to acquire a mean value of 0, standard deviation is 1, the source modal images after processing are inputted, prediction is carried out through a convolutional neural network, a convolutional neural network prediction result is outputted to carry out linear transformation to generate a target modal image; a purpose of linear transformation is realized, and the target modal image is generated. Medical image modal synthesis can be rapidly and effectively realized through the convolutional neural network acquired through learning, different-scale and multi-layer image characteristics can be automatically learned and optimized through training the convolutional neural network, a time-consuming image registering process is not needed, and rapid and accurate modal synthesis can be accomplished.

Description

technical field [0001] The invention relates to the technical field of digital image processing, in particular to a method for modality synthesis of medical images. Background technique [0002] Synthesis and virtual reconstruction of corresponding images through computer simulation of another imaging device's imaging method, this process is called modality synthesis, such as reconstruction of ultrasound images from X-ray computed tomography image data, images obtained by magnetic resonance imaging equipment Data prediction and synthesis of corresponding CT images, etc. The purpose of the present invention is to provide an effective modality synthesis method to realize modality synthesis of medical images quickly and effectively. The results generated by modality synthesis can be used to simplify multimodal medical image registration problems, attenuation correction in PET / MRI imaging systems, radiation dose distribution calculation in radiotherapy, etc. [0003] Ordinary ...

AI Technical Summary

Problems solved by technology

[0003] Ordinary digital image processing methods are difficult to apply to the image data acquired by most existing imaging equipment and conventional imaging methods when establishing physical models, and have limited universality and narrow application range. However, the atlas method is applicable to limited parts, and the accuracy of modal synthetic images is limited. It cannot be guaranteed that all individual images correspond well to the atlas, and atlas registration is very time-consuming. Establishing a regression prediction model can only extract a small The image features of the region are difficult to approximate the complex mapping relationship between different modalities of medical images

Images