A three-dimensional segmentation method of brain MRI hippocampus based on deep learning

Abstract

The present invention relates to the fields of computer vision and machine learning, and in particular to a three-dimensional segmentation method of brain MRI hippocampus based on deep learning. The steps are as follows: step 1, preprocessing the original image set A; step 2, building a network model; The three-dimensional segmentation network model of brain MRI hippocampus includes 3 one-dimensional convolutional layers, 15 three-dimensional convolutional layers and 4 maximum pooling layers. The entire network model is divided into left and right parts. To capture the content of the image, the expansion path on the right is used for precise positioning. On the left is a downsampling process, which is divided into five groups of convolution operations. Step 3, training the model; using the standardized image set E as the training set to train the three-dimensional segmentation network model of the brain MRI hippocampus in step 2, and obtaining the trained network model F for the three-dimensional segmentation of the brain MRI hippocampus. While ensuring high segmentation precision, the invention has fast operation speed and strong expandability.

Description

technical field [0001] The present invention relates to the fields of computer vision and machine learning, in particular to a three-dimensional segmentation method of brain MRI hippocampus based on deep learning. Background technique [0002] The early clinical manifestations of Alzheimer's disease (commonly known as senile dementia) are hippocampal atrophy. Doctors can perform three-dimensional imaging of the patient's brain through nuclear magnetic resonance technology, and then conduct diagnosis and design related treatment plans based on image analysis. When judging whether the hippocampus is shrinking, doctors usually need to segment the hippocampal structure and perform shape and volume analysis. However, the size of the hippocampus is small, the shape is irregular and varies from person to person, and under conventional MRI images, the contrast with the surrounding tissue structure is low, and the boundary is unclear or even discontinuous. It is difficult for radiol...

AI Technical Summary

Problems solved by technology

However, the size of the hippocampus is small, the shape is irregular and varies from person to person, and the contrast with the surrounding tissue structure is low under conventional MRI images, and the boundary is unclear or even discontinuous

It is difficult for radiologists with many years of clinical experience to perform accurate segmentation

[0003] However, the ratio of doctors to patients in my country is extremely disparate, and the scarce doctor resources are far from being able to meet the needs of the huge patient population.

Moreover, the medical strength of grass-roots hospitals is weak, and the level of doctors is uneven, causing a large number of patients to flock to large tertiary hospitals, further exacerbating the imbalance of doctor-patient ratio

Images