3d医学图像数据载入（）torchio库

 import一堆东西先（不一定全用上）

from glob import glob
from os.path import dirname, join, basename, isfile
import sys
sys.path.append('./')
import csv
import torch
from medpy.io import load
import numpy as np
from PIL import Image
from torch import nn
import torch.nn.functional as F
import random
import torchio as tio
import nibabel as nib
import numpy as np
import torchio as tio
from torchio import AFFINE, DATA
import torchio
from torchio import ScalarImage, LabelMap, Subject, SubjectsDataset, Queue
from torchio.data import UniformSampler
from torch.utils.data import DataLoader
from torchio.transforms import (
    RandomFlip,
    RandomAffine,
    RandomElasticDeformation,
    RandomNoise,
    RandomMotion,
    RandomBiasField,
    RescaleIntensity,
    Resample,
    ToCanonical,
    ZNormalization,
    CropOrPad,
    HistogramStandardization,
    OneOf,
    Compose,
)
from pathlib import Path

from hparam import hparams as hp

设定路径

source_train_0 = r'C:\Users\Admin\SIAT\model\Pytorch-Medical-Classification-main\0'
source_train_1 = r'C:\Users\Admin\SIAT\model\Pytorch-Medical-Classification-main\1'

 读取文件

class MedData_train(torch.utils.data.Dataset):
    def __init__(self, images_dir_0, images_dir_1):

        self.subjects = []


        images_dir_0 = Path(images_dir_0)
        self.image_paths_0 = sorted(images_dir_0.glob(hp.fold_arch))

        images_dir_1 = Path(images_dir_1)
        self.image_paths_1 = sorted(images_dir_1.glob(hp.fold_arch))

        for (image_path) in zip(self.image_paths_0):
            subject = tio.Subject(
                source=tio.ScalarImage(image_path),
                label= 0,
            )
            self.subjects.append(subject)

        for (image_path) in zip(self.image_paths_1):
            subject = tio.Subject(
                source=tio.ScalarImage(image_path),
                label= 1,
            )
            self.subjects.append(subject)

        self.transforms = self.transform()

        self.training_set = tio.SubjectsDataset(self.subjects, transform=self.transforms)


        # one_subject = self.training_set[0]
        # one_subject.plot()

 3D医学图像裁剪、归一增强等处理

    def transform(self):

        if hp.aug: #数据增强开关
            training_transform = Compose([
            CropOrPad((hp.crop_or_pad_size), padding_mode='reflect'),#切割跟扩充
            # ToCanonical(),#规范坐标轴
            # RandomBiasField(),#随机 MRI 偏置场伪影
            ZNormalization(),#正则化
            # RandomNoise(),#高斯噪声

            # RandomGhosting(),#MRI 鬼影伪影
            # RandomMotion(),#运动伪影
            # RandomSpike(), #MRI 尖峰伪影
            # RandomBlur(),#随机大小的高斯滤波器模糊图像
            # RandomSwap(),#随机交换图像中的补丁


            RandomFlip(axes=(0,)),#反转图像中元素的顺序
            RandomAffine(),

            # OneOf({
            #     # RandomAffine(): 0.3,#倾斜
            #     # RandomElasticDeformation(): 0.1,#随机弹性变形
            # }),
            ])


        return training_transform

 载入

def get_load():
    # dataset_ = PET_dataset(dir)
    # data_loader = DataLoader(dataset=dataset_,batch_size=batch_size,shuffle=shuffle,num_workers=num_workers)
    train_dataset = MedData_train(source_train_0, source_train_1)
    train_loader = DataLoader(train_dataset.training_set,
                              batch_size=1,
                              shuffle=True,
                              pin_memory=True,
                              drop_last=True)
    return train_loader

 测试

train_loader = get_load()

for i, batch in enumerate(train_loader):
    x = batch['source']['data']
    # imag = x.reshape([64,64,64,1]) #将之转化为数组查看
    # imag = imag.numpy()
    y = batch['label']

    x = x.type(torch.FloatTensor).cuda()
    y = y.type(torch.LongTensor).cuda()
    print(x.shape)
    print(y)

    # img_t1 = nib.Nifti1Image(imag, np.eye(4))
    # nib.save(img_t1, 'output.nii.gz')#将之保存为nii查看
    # print(imag.shape)