LIDC-IDRI数据集

ChemXL

已于 2024-01-30 13:16:06 修改

阅读量1k

点赞数 8

文章标签：目标检测计算机视觉深度学习神经网络人工智能图像处理 opencv

于 2024-01-30 13:05:51 首次发布

本文链接：https://blog.csdn.net/cxldbbv/article/details/135930575

版权

LIDC-IDRI数据集xml文件读取roi坐标，并根据imageSOP_UID匹配到原始dcm图像上框出肺结节区域

import pydicom as dicom
import os
import cv2
import numpy as np
import lxml.etree as etree
import pydicom
from PIL import Image
from skimage.draw import polygon
from matplotlib.pyplot import plot as plt


# 这个函数输入的是xml的路径，返回的是读取的坐标
# 这里获得的坐标是所有层的，但只读取了单个病理医生标注的第一个结节，后面的全都不读，要是想要读的话可以改一下代码
def xml_reader(path):
    anno = {}
    xmlns = '{http://www.nih.gov}'
    tree = etree.parse(path)
    # for ann in tree.findall('./Annotation'):
    assert tree.findall(xmlns + 'ResponseHeader') != [], '{}.has no ResponseHeader'.format(path)
    Res = tree.findall(xmlns + 'ResponseHeader')[0]
    # anno['SeriesInstanceUid']   = Res.findall(xmlns + 'SeriesInstanceUid')[0].text
    # anno['StudyInstanceUID']    = Res.findall(xmlns + 'StudyInstanceUID')[0].text
    # anno['session'] = []
    if tree.findall(xmlns + 'readingSession') == []: return {}
    i = tree.findall(xmlns + 'readingSession')[0]
    Nodules = []
    if i.findall(xmlns + 'unblindedReadNodule') == []: return {}
    j = i.findall(xmlns + 'unblindedReadNodule')[0]
    Nodule = []
    for k in j.findall(xmlns + 'roi'):
        roi = {}
        roi['imageSOP_UID'] = k.findall(xmlns + 'imageSOP_UID')[0].text
        # cxl = roi['imageSOP_UID'].split('\n')
        # print(type(cxl))
        # print(roi)
        roi['roi'] = []
        for l in k.findall(xmlns + 'edgeMap'):
            x = int(l.findall(xmlns + 'xCoord')[0].text)
            y = int(l.findall(xmlns + 'yCoord')[0].text)
            roi['roi'].append([x, y])
        #print(roi)
        Nodule.append(roi)
    #print(Nodule)
    # for i in Nodule:
    #     print(i)
    Nodules.append(Nodule)
    anno['session'] = Nodules
    return anno

xml_name = 'E:\manifest-1600709154662\LIDC-IDRI\LIDC-IDRI-0001\\01-01-2000-NA-NA-30178\\3000566.000000-NA-03192\\069.xml'
roi_contour = xml_reader(xml_name)
# 获取读取到的roi的坐标点
r = roi_contour['session'][0]
# print(r)
#print(r[0]['roi'])

#查找切片对应的dcm图片

# 假设你已经从XML中获取到了目标SOP Instance UID
target_sop_uid = r[0]['imageSOP_UID']
# DICOM图像所在目录路径
dcm_folder_path = "D:\Paper_Code\Test_medical_image\manifest-1600709154662\LIDC-IDRI\LIDC-IDRI-0001\\01-01-2000-NA-NA-30178\\3000566.000000-NA-03192"

cxl = ''
# 遍历DICOM文件夹，查找匹配的SOP Instance UID
for root, dirs, files in os.walk(dcm_folder_path):
    for file_name in files:
        if file_name.endswith(".dcm"):
            # 打开DICOM文件并读取SOP Instance UID
            dcm_file_path = os.path.join(root, file_name)
            ds = pydicom.dcmread(dcm_file_path)

            # 比较当前文件的SOP Instance UID是否与目标一致
            if ds.SOPInstanceUID == target_sop_uid:
                print(f"找到了匹配的DICOM切片：{dcm_file_path}")
                cxl = dcm_file_path
                # 这里可以进一步处理找到的DICOM文件
                break

#打开对应的dcm图像

# 打开DICOM文件
ds = pydicom.dcmread(cxl)
# 获取图像数据
image_data = ds.pixel_array

#填充结节区域
temp = np.array(image_data, np.uint8)
temp = cv2.cvtColor(temp, cv2.COLOR_GRAY2RGB)
mask = np.array(r[7]['roi'], dtype=np.int32)
# # 用cv2根据多边形画图。
# temp = cv2.fillConvexPoly(temp, mask, (255, 255, 255))
# cv2.imshow("ROI Mask", temp)
# cv2.waitKey(0)
# cv2.destroyAllWindows()

# 描绘结节边界
color = (0, 0, 255)  # 白色
thickness = 2             # 边界线宽
is_closed = True          # 是否闭合轮廓

# 使用cv2.polylines()绘制边界
cv2.polylines(temp, [mask], is_closed, color, thickness)

# 显示图像
cv2.imshow("ROI Boundary", temp)
cv2.waitKey(0)
cv2.destroyAllWindows()