import csv
import platform
import numpy as np
import matplotlib.pyplot as plt
import os
import otsu
from FeatureExtract import *
from skimage.feature import hog, haar_like_feature,local_binary_pattern,multiblock_lbp,daisy
import pysift
import time
def readCSV(filename):
lines = []
with open(filename, "r") as f:
csvreader = csv.reader(f)
for line in csvreader:
lines.append(line)
return lines
ori_path = 'C:/Users/RL/Desktop/可解释性的特征学习分类/nodules/ori_hu/'
def get_feature_name(temp, index, value):
name = []
for t in temp:
if float(t[index]) == value:
if len(t[1]) == 1:
name.append('LIDC-IDRI-000' + t[1] + '_' + t[3] + '_' + t[2] + '.npy')
elif len(t[1]) == 2:
name.append('LIDC-IDRI-00' + t[1] + '_' + t[3] + '_' + t[2] + '.npy')
elif len(t[1]) == 3:
name.append('LIDC-IDRI-0' + t[1] + '_' + t[3] + '_' + t[2] + '.npy')
else:
name.append('LIDC-IDRI-' + t[1] + '_' + t[3] + '_' + t[2] + '.npy')
return name
def read_name():
labelCSV = readCSV('C:/Users/RL/Desktop/可解释性的特征学习分类/特征图片/malignancy.csv')
Max = []
Min = []
temp = [labelCSV[i] for i in range(len(labelCSV))]
final = np.array([[float(temp[i][j + 21]) for j in range(9)] for i in range(len(temp)) if '0' not in temp[i]])
Max = final.max(axis=0)
Min = final.min(axis=0)
# prob_map = [[final[i][j] / (Max[j] + Min[j] + 1) if j != 8 else final[i][j] >= 3.5 for j in range(final.shape[1])] for i in range(final.shape[0])]
# 29 28 27毛刺 26分叶 25 24 23 22内在的 21
# print(prob_map[0])
name_maoci= get_feature_name(temp, 27, Max[6])
name_fenye = get_feature_name(temp, 26, Max[5])
name_solid = get_feature_name(temp, 28, Max[7])
name_non_solid = get_feature_name(temp, 28, 3)
name_moboli = get_feature_name(temp, 28, 1)
# with open('C:/Users/RL/Desktop/可解释性的特征学习分类/特征图片/name.txt',"w") as f:
# f.write("*" * 10 + "maoci:"+ '\n')
# for name in name_maoci:
# f.write(name + '\n')
# f.write("*" * 10 + "fenye:"+ '\n')
# for name in name_fenye:
# f.write(name+ '\n')
# f.write("*" * 10 + "shixing:"+ '\n')
# for name in name_solid:
# f.write(name+ '\n')
# f.write("*" * 10 + "yashixing:"+ '\n')
# for name in name_non_solid:
# f.write(name+ '\n')
# f.write("*" * 10 + "moboli:"+ '\n')
# for name in name_moboli:
# f.write(name+ '\n')
return name_maoci, name_fenye, name_solid, name_non_solid, name_moboli
def read_lidc(filename):
image = np.load(ori_path + filename)
return image
def image_feature_extract(image):
features = [image]
# otsu_image = otsu.helper(image.copy())
features.append(otsu.helper(image.copy()))
features.append(gabor(image.copy()))
# 添加新的方式
# lbp = local_binary_pattern(image.copy(), 3, 3, method='var')
# features.append(otsu._otsu(lbp))
# features.append(edge_detection(image.copy()))
_, hog_image = hog(gabor(image.copy()), orientations=9, pixels_per_cell=(8, 8),cells_per_block=(2, 2), visualize=True, multichannel=False)
features.append(hog_image)
features.append(np.transpose(super_pixel(image.copy()),(2,0,1))[0])
features.append(local_binary_pattern(image.copy(), 4, 4, method='var')) # 会产生无穷大或者是无穷小
fd, hog_image = hog(image.copy(), orientations=9, pixels_per_cell=(4, 4),cells_per_block=(2, 2), visualize=True, multichannel=False)
_, descs_img = daisy(image.copy(), step=180, radius=58, rings = 2, histograms=10,orientations=8,visualize=True)
features.append(hog_image)
features.append(np.transpose(descs_img.copy(),(2,0,1))[0])
features.append(descs_img)
kps, _ = pysift.computeKeypointsAndDescriptors(image.copy())
x = []
y = []
for kp in kps:
x.append(kp.pt[0])
y.append(kp.pt[1])
return features, x, y
# 接下里就是 统计每个属性的图像特征输出
if __name__ == "__main__":
a1, a2, a3, a4, a5 = read_name()
ori_image = []
# 不清楚normalize的影响
ori_image.append([normalization(truncate_hu(read_lidc('LIDC-IDRI-0007_2_3000631.npy')))])# 毛刺
ori_image.append([normalization(truncate_hu(read_lidc('LIDC-IDRI-0060_1_3000575.npy')))])# 分叶
ori_image.append([normalization(truncate_hu(read_lidc('LIDC-IDRI-0003_4_3000611.npy')))]) # 磨玻璃
ori_image.append([normalization(truncate_hu(read_lidc('LIDC-IDRI-0003_2_3000611.npy')))]) # 实性
ori_image.append([normalization(truncate_hu(read_lidc('LIDC-IDRI-0008_1_3000549.npy')))]) # 亚实性
ori_image.append([normalization(truncate_hu(read_lidc('LIDC-IDRI-0132_1_5418.npy')))]) # 空洞
X = []
Y = []
for i in range(6):
features, x, y = image_feature_extract(ori_image[i][0])
for feature in features:
ori_image[i].append(feature)
# print(len(ori_image[i]))
X.append(x)
Y.append(y)
name = ["origial","sift-key-point","otsu","gabor","new","super-pixel","lbp","hog","daisy-three-dim","daisy-gray"]
plt.figure()
# 对实性结节 对亚实性结节 对分叶 对毛玻璃 对空洞 对毛刺
numRows = 6
numCols = 10
font2 = {'family' : 'Times New Roman',
'weight' : 'normal',
'size' : 7,
}
for i in range(numRows):
for j in range(numCols):
ax = plt.subplot(numRows,numCols,1 + i * numCols + j)
if i == 0:
ax.set_title(name[j],font2)
if j != numCols - 1:
plt.imshow(ori_image[i][j], cmap="gray")
else:
plt.imshow(ori_image[i][j])
if j == 1:
plt.scatter(X[i], Y[i], color='red', s=3, alpha=0.5)
plt.xticks([])
plt.yticks([])
# plt.savefig("C:/Users/RL/Desktop/可解释性的特征学习分类/特征图片/pictures/test" + str(time.time()) + ".png",dpi=1500)
plt.show()
Datasets
Models
