Heart nuclear magnetic resonance image key point detection method based on convolutional neural network

Abstract

The invention discloses a heart nuclear magnetic resonance image key point detection method based on a convolutional neural network, and the method comprises the steps of constructing the convolutional neural network, efficiently detecting the regions of a plurality of key points through a pixel-by-pixel classifier, predicting the distances from pixels to the centers of the key points, and guaranteeing the precise prediction of the key points. The false positive rate of detection is reduced and the accuracy of key point detection is ensured by removing false detection key points, removing leftand right ventricular intersection points, which are not between the left ventricle and the right ventricle, in the fusion information graph and obtaining the local maximum value in the fusion information graph according to the local maximum value search method.

Description

technical field [0001] This field belongs to the field of key point detection of cardiac nuclear magnetic resonance images, and in particular relates to a method for detecting key points of cardiac nuclear magnetic resonance images based on a convolutional neural network. Background technique [0002] As tomography (CT) and magnetic resonance imaging (MR) are widely used in disease diagnosis, treatment planning and clinical research, medical image computer-aided diagnosis (CAD) has become a part of the daily work of doctors in clinical diagnosis and treatment planning. An important step of the technology, and anatomical structure key point detection is an important research hotspot of this technology. Key point detection technology can help doctors quickly locate lesions, organs and other objects of interest, improving diagnostic efficiency. The key point detection algorithm has achieved good detection results in medical image processing, but it still has certain limitation...

AI Technical Summary

Problems solved by technology

In the prior art, for threshold-based methods, morphological processing-based and shape feature-based detection methods, noise interference in MRI images, uneven distribution of pixel values, differences in the shape of the left ventricle, and high similarity between the ventricle and surrounding tissues are all problems. It brings great challenges to the robustness of the detection algorithm

In addition, the hybrid model of sparse autoencoder and SVM classifier does not have end-to-end characteristics, which affects the detection efficiency

Images