【ADNI】数据预处理（6）ADNI_slice_dataloader ||| show image
ADNI Series
1、【ADNI】数据预处理（1）SPM，CAT12
2、【ADNI】数据预处理（2）获取 subject slices
3、【ADNI】数据预处理（3）CNNs
4、【ADNI】数据预处理（4）Get top k slices according to CNNs
5、【ADNI】数据预处理（5）Get top k slices (pMCI_sMCI) according to CNNs
6、【ADNI】数据预处理（6）ADNI_slice_dataloader ||| show image
本文总结的功能：
1）将来自同一个 subject MRI image 的 slice concatenate 到同一数组；
2）显示 slices image；
subject ID and relevant label：存储与 .txt 文本中，如下所示
141_S_1137 1
141_S_1152 1
002_S_0295 0
002_S_0559 0subject id ||| slice path ||| label name：
141_S_0696|||/home/hcq/alzheimer_disease/ADNI_825/experiments_FineTunning/majority_select_top51_slices_folder_02_AD_NC/validation/AD/141_S_0696_slice_Y74.jpg|||AD
141_S_0696|||/home/hcq/alzheimer_disease/ADNI_825/experiments_FineTunning/majority_select_top51_slices_folder_02_AD_NC/validation/AD/141_S_0696_slice_Z46.jpg|||AD
002_S_0413|||/home/hcq/alzheimer_disease/ADNI_825/experiments_FineTunning/majority_select_top51_slices_folder_02_AD_NC/validation/NC/002_S_0413_slice_X46.jpg|||NC
002_S_0413|||/home/hcq/alzheimer_disease/ADNI_825/experiments_FineTunning/majority_select_top51_slices_folder_02_AD_NC/validation/NC/002_S_0413_slice_X50.jpg|||NC
效果：
1）subject：image = [batch_size, slice_num, img_w, img_h]
2）label：AD or NC
源码：
import random
import os
import numpy as np
import matplotlib.image as mpimg
import matplotlib.pyplot as plt
import cv2
def read_lists(file_path):
dataset_list = np.loadtxt(file_path, dtype=str).tolist()
subjectID_path, labels = zip(*[(l[0], int(l[1])) for l in dataset_list])
return subjectID_path, labels
class Dataset_ADNI_TopK:
def __init__(self, subjectID_list, subjectID_labels, folder_name):
self.subjectID_list = subjectID_list
self.subjectID_labels = subjectID_labels
self.slice_path = os.path.join("./path_txt", folder_name + "_majority_select_top51_slices_folder_02.txt")
self.shuffled = False
def slice_concatenate(self, batch_size, slice_num):
# for subjectID in self.subjectID_list:
# print(subjectID)
slice_list = np.loadtxt(self.slice_path, dtype=str).tolist()
x = np.zeros((batch_size, slice_num, 227, 227))
y = np.zeros(batch_size)
for bs in range(batch_size):
subjectID = self.subjectID_list[bs]
subjectID_label = self.subjectID_labels[bs]
# print(subjectID)
for i, slice_strcut in enumerate(slice_list):
# print(slice_strcut.split("|||")[1])
image = mpimg.imread(slice_strcut.split("|||")[1])
# image = np.reshape(image, (227, 227))
image = cv2.resize(image, (227, 227))
# print(image.shape)
# print(i%slice_num)
x[bs, i%slice_num, :, :] = image
y[bs] = subjectID_label
# print(x[0, 0, 0, 90:100, 90:100])
return x, y
train_subject_ID = './path_txt/train_sujectID_majority_select_top51_slices_folder_02.txt'
val_subject_ID = './path_txt/val_sujectID_majority_select_top51_slices_folder_02.txt'
train_subject_ID_list, train_subject_ID_label_list = read_lists(train_subject_ID)
val_subject_ID_list, val_subject_ID_lable_list = read_lists(val_subject_ID)
# dataset_train = Dataset_ADNI_TopK(train_subject_ID_list, train_subject_ID_label_list, "train")
# dataset_train.slice_concatenate()
dataset_val = Dataset_ADNI_TopK(val_subject_ID_list, val_subject_ID_lable_list, "val")
batch_size_subject = 2
slice_num = 51
image, label = dataset_val.slice_concatenate(batch_size_subject, slice_num)
### show image
for i in range(16):
plt.subplot(4, 4, i+1)
plt.imshow(image[0, i, :, :], cmap = 'gray')
#plt.imshow(image[0, 0, :, :], cmap = 'gray')
#i = 1
#plt.subplot(2, 2, i)
#plt.imshow(image[0, 0, :, :])
#
#plt.subplot(222)
#plt.imshow(image[0, 1, :, :])
#
#plt.subplot(223)
#plt.imshow(image[0, 2, :, :])
#
#plt.subplot(224)
#plt.imshow(image[0, 3, :, :])hcq_data_processing.py
# -*- coding: utf-8 -*-
import random
import os
import numpy as np
import matplotlib.image as mpimg
import matplotlib.pyplot as plt
import cv2
def read_lists(file_path):
dataset_list = np.loadtxt(file_path, dtype=str).tolist()
subjectID_path, labels = zip(*[(l.split("|||")[0], int(l.split("|||")[1])) for l in dataset_list])
return subjectID_path, labels
class Dataset_ADNI_TopK:
def __init__(self, subjectID_list, subjectID_labels, folder_name, shuffled, batch_size, slice_num):
self.subjectID_list = subjectID_list
self.subjectID_labels = subjectID_labels
self.shuffled = shuffled
self.batch_size = batch_size
self.len_dataset = len(self.subjectID_list)
self.slice_num = slice_num
self.folder_name = folder_name
self.batch_size_used = batch_size
if shuffled == True:
# len_subjectID_list = len(subjectID_list)
# random.shuffle(len_subjectID_list)
# print("original list...")
# print(self.subjectID_list)
z = list(zip(self.subjectID_list, self.subjectID_labels))
random.shuffle(z)
self.subjectID_list, self.subjectID_labels = [list(l) for l in zip(*z)]
# print("random list...")
# print(self.subjectID_list)
def slice_concatenate(self, index):
# for subjectID in self.subjectID_list:
# print(subjectID)
root_path = "/home/reserch/documents/deeplearning/alzheimers_disease_DL/pytorch/dataset_path/original_825_top_entropy51"
model_img_size = 224
x = np.zeros((self.batch_size, self.slice_num, model_img_size, model_img_size))
y = np.zeros(self.batch_size)
for bs in range(self.batch_size):
if (index*self.batch_size + bs) >= self.len_dataset:
# print("break....")
# ### hcq ###
# ### 20180528 ###
# ### dataloader ###
# batch_size = 16
# train_dataset_num = 344 --> iter_num = train_dataset_num / batch_size = 21
# validation_dataset_num = 86 --> iter_num = validation_dataset_num / batch_size = 5
# # As for train_dataset_num:
# - iter_num = train_dataset_num / batch_size = 21
# - range(iter_num)[0, 20]: for _iter in iter_num: dealed with = 21 iterations --> 21 x 16 = 336
# - remained samples: 344 - 336 = 8 <-- train_dataset_num % batch_size
# - range(iter_num+1)[0, 21]:
delete_samples_num = self.batch_size - (self.len_dataset % self.batch_size)
for ii in range(delete_samples_num):
index_delete = self.batch_size - ii - 1
x = np.delete(x, index_delete, axis=0)
y = np.delete(y, index_delete, axis=0)
self.batch_size_used = bs
break
# print("index = {}, dataset_len = {} ||| index*self.batch_size + bs = {}".format(index, self.len_dataset, index*self.batch_size + bs))
subjectID = self.subjectID_list[index*self.batch_size + bs]
subjectID_label = self.subjectID_labels[index*self.batch_size + bs]
y[bs] = subjectID_label
# print(subjectID)
new_slice_txt_path = os.path.join(root_path, self.folder_name + "_slice_txt", subjectID + "_" + self.folder_name + ".txt")
slice_list = np.loadtxt(new_slice_txt_path, dtype=str).tolist()
for i, slice_strcut in enumerate(slice_list):
# print(slice_strcut.split("|||")[1])
image = mpimg.imread(slice_strcut.split("|||")[1])
# image = np.reshape(image, (227, 227)) ## Not work
image = cv2.resize(image, (model_img_size, model_img_size))
# print(image.shape)
# print(i%slice_num)
x[bs, i%self.slice_num, :, :] = image
## return a batch_size pair(images, labels)
# print("x.shape = {}, y.shape = {}".format(x.shape, y.shape))
return x, y
def iter_len(self):
iter_num = self.len_dataset / self.batch_size
# print("iter_len = {} ||| len_dataset = {}, batch_size = {}".format(num, self.len_dataset, self.batch_size))
return iter_num
#train_subject_ID = './path_txt/train_sujectID_majority_select_top51_slices_folder_02.txt'
#val_subject_ID = './path_txt/val_sujectID_majority_select_top51_slices_folder_02.txt'
#train_subject_ID_list, train_subject_ID_label_list = read_lists(train_subject_ID)
#val_subject_ID_list, val_subject_ID_lable_list = read_lists(val_subject_ID)
#
## dataset_train = Dataset_ADNI_TopK(train_subject_ID_list, train_subject_ID_label_list, "train")
## dataset_train.slice_concatenate()
#
#batch_size_subject = 1
#dataset_val = Dataset_ADNI_TopK(val_subject_ID_list, val_subject_ID_lable_list, "val", True, batch_size_subject)
#slice_num = 51
#image, label = dataset_val.slice_concatenate(slice_num)
#dataset_val.iter_len()Reference：
- PyTorch ImageFolder：https://github.com/pytorch/vision/blob/master/torchvision/datasets/folder.py