MR 的几种归一化方法总结

1、nMax = data.max() nMin = data.min() nDis = nMax-nMin data = (data-nMin)/nDis

Refimg = sitk.ReadImage(os.path.join(NII_DIR, 'first_follow_up_T1.nii.gz'))
data = sitk.GetArrayFromImage(Refimg)
nMax = data.max()
nMin = data.min()
nDis = nMax-nMin
data = (data-nMin)/nDis

#normalize_data_storage(data)
out = sitk.GetImageFromArray(data)
out.SetSpacing(Refimg.GetSpacing())
out.SetOrigin(Refimg.GetOrigin())
sitk.WriteImage(out, os.path.join(NII_DIR, 'first_baseline_NCCT.nii.gz'))

2、data -= mean[:, np.newaxis, np.newaxis, np.newaxis] data /= std[:, np.newaxis, np.newaxis, np.newaxis]

def normalize_data(data, mean, std):
    # data：[4,144,144,144]
    data -= mean[:, np.newaxis, np.newaxis, np.newaxis]
    data /= std[:, np.newaxis, np.newaxis, np.newaxis]
    return data


def normalize_data_storage(data_storage):
    means = list()
    stds = list()
    # [n_example,4,144,144,144]
    for index in range(data_storage.shape[0]):
        # [4,144,144,144]
        data = data_storage[index]
        # 分别求出每个模态的均值和标准差
        means.append(data.mean(axis=(1, 2, 3)))
        stds.append(data.std(axis=(1, 2, 3)))
    # 求每个模态在所有样本上的均值和标准差[n_example,4]==>[4]
    mean = np.asarray(means).mean(axis=0)
    std = np.asarray(stds).mean(axis=0)
    for index in range(data_storage.shape[0]):
        # 根据均值和标准差对每一个样本归一化
        data_storage[index] = normalize_data(data_storage[index], mean, std)
    return data_storage