Three-dimensional nuclear magnetic resonance pancreatic image segmentation method based on multi-step learning

Abstract

The invention discloses a three-dimensional nuclear magnetic resonance pancreas image segmentation method based on multi-step learning. The method comprises the following steps: (1) carrying out normalization preprocessing on a three-dimensional nuclear magnetic resonance image; (2) randomly dividing the preprocessed images into a training set, a verification set and a test set; (3) segmenting thepancreas-containing blocks in the training set and the corresponding marks by using segmentation marks for later use; (4) compressing the original image to train a Q-image; the net model calculates the approximate position of the pancreas; (5) pre-training another P-net model by using the pairing data in the step 3; (6) using pre-trained Q-Net generates a 3D position prediction graph of pancreas,selects a graph block with high probability, maps the graph block back to an original graph, and inputs pre-trained P-net in blocks; performing combined training in net to predict the pancreas position; and (7) predicting the detection effect on the test set by using the trained segmentation model. By utilizing the method, the pancreas can be accurately segmented from the three-dimensional nuclear magnetic resonance image, and a basis and guidance can be provided for the radiation treatment of the pancreas.

Description

Technical field [0001] The invention belongs to the field of medical image processing, and particularly relates to a three-dimensional nuclear magnetic resonance pancreatic image segmentation method based on multi-step learning. Background technique [0002] Pancreatic cancer is a kind of cancer that appears on the pancreas that has been a serious threat to human health for a long time. It is the cancer with the highest degree of malignancy and the highest mortality among common malignant tumors, and its incidence and mortality are almost equal. About 90% of the patients cannot be cured by surgery, the metastasis is extremely fast, and the five-year survival rate is only 3%. Although with the continuous progress and development of science and technology, the overall treatment of cancer has made great breakthroughs, but the treatment of pancreatic cancer has not made substantial progress, and the mortality rate has also ranked first for a long time. Therefore, in the diagnosis an...

AI Technical Summary

Problems solved by technology

[0006]In the existing pancreas segmentation methods, such as region-based segmentation, edge-based segmentation, map-based segmentation, etc., are all based on two-dimensional sequential image segmentation methods , using the slice information of the 3D nuclear magnetic resonance image for segmentation, the relevant information between different slices cannot be effectively used

On the other hand, due to the inherent characteristics of the pancreas, the volume is too small compared to the abdominal cavity, and the difference in the number of positive and negative samples is too large. It is difficult to accurately segment the location of the pancreas using the existing direct segmentation method. At the same time, MRI images have the characteristics of high noise, low contrast, and unclear borders of adjacent tissues, which makes this task difficult to perform through ordinary 3D segmentation neural networks, often resulting in blurred boundaries, under-segmentation or over-segmentation

Images