SimpleITK三维图像分析

SimpleITK三维图像分析

  • 1、去除3D 小连通域
  • 2、【医学图像处理】之腹部骨骼提取(SimpleITK)

1、去除3D 小连通域

在一些计算机视觉任务中,需要对模型的输出做一些后处理以优化视觉效果,连通域就是一种常见的后处理方式。尤其对于分割任务,有时的输出mask会存在一些假阳(小的无用轮廓),通过3D连通域找出面积较小的独立轮廓并去除可以有效地提升视觉效果。

二维图像连通域一般包括 4连通、8连通。对于三维数据一般包括6连通、18连通和26联通。
下面的代码只保留最大3D连通域。

# -*- coding : UTF-8 -*-
# @file   : prob2label.py
# @Time   : 2021-10-19 9:35
# @Author : wmz

import os
import SimpleITK as sitk
from glob import glob
import numpy as np


def getFiles(path, suffix):
    return [os.path.join(root, file) for root, dirs, files in os.walk(path) for file in files if file.endswith(suffix)]


def connected_domain_2(image, mask=True):
    cca = sitk.ConnectedComponentImageFilter()
    cca.SetFullyConnected(True)
    _input = sitk.GetImageFromArray(image.astype(np.uint8))
    output_ex = cca.Execute(_input)
    stats = sitk.LabelShapeStatisticsImageFilter()
    stats.Execute(output_ex)
    num_label = cca.GetObjectCount()
    num_list = [i for i in range(1, num_label+1)]
    area_list = []
    for l in range(1, num_label +1):
        area_list.append(stats.GetNumberOfPixels(l))
    num_list_sorted = sorted(num_list, key=lambda x: area_list[x-1])[::-1]
    largest_area = area_list[num_list_sorted[0] - 1]
    final_label_list = [num_list_sorted[0]]

    # for idx, i in enumerate(num_list_sorted[1:]):  # 大于第一个的十分之一的都保留,注释掉之后只保留最大连通域
    #     if area_list[i-1] >= (largest_area//10):
    #         final_label_list.append(i)
    #     else:
    #         break
    output = sitk.GetArrayFromImage(output_ex)

    for one_label in num_list:
        if one_label in final_label_list:
            continue
        x, y, z, w, h, d = stats.GetBoundingBox(one_label)
        one_mask = (output[z: z + d, y: y + h, x: x + w] != one_label)
        output[z: z + d, y: y + h, x: x + w] *= one_mask

    if mask:
        output = (output > 0).astype(np.uint8)
    else:
        output = ((output > 0)*255.).astype(np.uint8)
    return output


def save_prob2label(prob_dir, save_labeldir):
    # all_prob_seg = glob(os.path.join(prob_dir, "*.nrrd"))
    all_prob_seg = getFiles(prob_dir, ".nrrd")
    for index, file in enumerate(all_prob_seg):
        print("processing", index + 1, '/', len(all_prob_seg), file)
        label_file = file.replace(prob_dir, save_labeldir).replace(".nrrd", ".nii.gz")
        prob_img = sitk.ReadImage(file)
        prob_arr = sitk.GetArrayFromImage(prob_img)
        label_arr = (prob_arr > Dice_value) * 1
        label_arr = connected_domain_2(label_arr)
        label_img = sitk.GetImageFromArray(label_arr)
        label_img.SetOrigin(prob_img.GetOrigin())
        label_img.SetDirection(prob_img.GetDirection())
        dst_dir = label_file.rsplit('\\', 1)[0]
        if not os.path.exists(dst_dir):
            os.makedirs(dst_dir)
        sitk.WriteImage(label_img, label_file)


if __name__ == '__main__':

    prob_nrrd_dir = r'C:\Users\wmz\Desktop\input'
    save_label_dir = r'C:\Users\wmz\Desktop\test'
    Dice_value = 0.5
    save_prob2label(prob_nrrd_dir, save_label_dir)


参考:python实现3D连通域后处理

2、【医学图像处理】之腹部骨骼提取(SimpleITK)

1.内容
步骤:
1.读取Dicom序列
2.设置固定阈值为100,把骨骼和心脏及主动脉都分割出来
3.形态学开运算+最大连通域提取,粗略的心脏和主动脉图像
4.将step1的结果与step2的结果相减,得到骨骼部分
5.最大连通域提取,去除小连接
6.将得到的图像与原始图像进行逻辑与操作
数据地址:
链接:https://pan.baidu.com/s/198H5g30LSKrKInJfgV1xFQ
提取码:a3nw

import SimpleITK as sitk

# 最大连通域提取
def GetLargestConnectedCompont(binarysitk_image):
    cc = sitk.ConnectedComponent(binarysitk_image)
    stats = sitk.LabelIntensityStatisticsImageFilter()
    stats.SetGlobalDefaultNumberOfThreads(8)
    stats.Execute(cc, binarysitk_image)
    maxlabel = 0
    maxsize = 0
    for l in stats.GetLabels():
        size = stats.GetPhysicalSize(l)
        if maxsize < size:
            maxlabel = l
            maxsize = size
    labelmaskimage = sitk.GetArrayFromImage(cc)
    outmask = labelmaskimage.copy()
    outmask[labelmaskimage == maxlabel] = 255
    outmask[labelmaskimage != maxlabel] = 0
    outmask_sitk = sitk.GetImageFromArray(outmask)
    outmask_sitk.SetDirection(binarysitk_image.GetDirection())
    outmask_sitk.SetSpacing(binarysitk_image.GetSpacing())
    outmask_sitk.SetOrigin(binarysitk_image.GetOrigin())
    return outmask_sitk

# 逻辑与操作
def GetMaskImage(sitk_src, sitk_mask, replacevalue=0):
    array_src = sitk.GetArrayFromImage(sitk_src)
    array_mask = sitk.GetArrayFromImage(sitk_mask)
    array_out = array_src.copy()
    array_out[array_mask == 0] = replacevalue
    outmask_sitk = sitk.GetImageFromArray(array_out)
    outmask_sitk.SetDirection(sitk_src.GetDirection())
    outmask_sitk.SetSpacing(sitk_src.GetSpacing())
    outmask_sitk.SetOrigin(sitk_src.GetOrigin())
    return outmask_sitk


# 读取Dicom序列
pathDicom = 'D:/PyCharm 2019.3.3/data/LIDC_nodul'
reader = sitk.ImageSeriesReader()
filenamesDICOM = reader.GetGDCMSeriesFileNames(pathDicom)
reader.SetFileNames(filenamesDICOM)
sitk_src = reader.Execute()

# step1.设置固定阈值为100,把骨骼和心脏及主动脉都分割出来
sitk_seg = sitk.BinaryThreshold(sitk_src, lowerThreshold=100, upperThreshold=3000, insideValue=255, outsideValue=0)
sitk.WriteImage(sitk_seg, 'step1.mha')

# step2.形态学开运算+最大连通域提取,粗略的心脏和主动脉图像
sitk_open = sitk.BinaryMorphologicalOpening(sitk_seg != 0, 2)
sitk_open = GetLargestConnectedCompont(sitk_open)
sitk.WriteImage(sitk_open, 'step2.mha')

# step3.再将step1的结果与step2的结果相减,得到骨骼部分
array_open = sitk.GetArrayFromImage(sitk_open)
array_seg = sitk.GetArrayFromImage(sitk_seg)
array_mask = array_seg - array_open
sitk_mask = sitk.GetImageFromArray(array_mask)
sitk_mask.SetDirection(sitk_seg.GetDirection())
sitk_mask.SetSpacing(sitk_seg.GetSpacing())
sitk_mask.SetOrigin(sitk_seg.GetOrigin())
sitk.WriteImage(sitk_mask, 'step3.mha')

# step4.最大连通域提取,去除小连接
skeleton_mask = GetLargestConnectedCompont(sitk_mask)
sitk.WriteImage(skeleton_mask, 'step4.mha')

# step5.将得到的图像与原始图像进行逻辑与操作
sitk_skeleton = GetMaskImage(sitk_src, skeleton_mask, replacevalue=-1500)
sitk.WriteImage(sitk_skeleton, 'step5.mha')

参考:【医学图像处理】之腹部骨骼提取(SimpleITK)

你可能感兴趣的:(ITK,ITK)