In [4]:
#Loading a test image and visualizing using nibabel package
img = nib.load('Caltech_0051487_reho.nii')
gg=img.get_fdata()
plotting.plot_img(img)
plt.show()
import pandas
import sklearn
import pwlf
import numpy as np
import matplotlib.pyplot as plt
import statsmodels.api as sm
import glob
import nibabel as nib
from tensorflow import keras
import tensorflow as tf
from sklearn.model_selection import train_test_split
from sklearn.model_selection import cross_val_score
from sklearn.model_selection import cross_validate
from sklearn.model_selection import KFold
from sklearn.ensemble import RandomForestClassifier
from nilearn import plotting
#5fold cross validation
crossvalidation = KFold(n_splits=5, random_state=1, shuffle=True)
#Loading a test image and visualizing using nibabel package
img = nib.load('Caltech_0051487_reho.nii')
gg=img.get_fdata()
plotting.plot_img(img)
plt.show()
# get data file names and raw data from images
import os
path1 =r'/Users/apoorvsara/Documents/autism/cpac/nofilt_noglobal/reho'
path2 =r'/Users/apoorvsara/Documents/controls/cpac/nofilt_noglobal/reho'
autism = [os.path.basename(x) for x in glob.glob('/Users/apoorvsara/Documents/autism/cpac/nofilt_noglobal/reho/**.nii.gz')]
controls = [os.path.basename(x) for x in glob.glob('/Users/apoorvsara/Documents/controls/cpac/nofilt_noglobal/reho/**.nii.gz')]
ocontrol_scans = np.array([nib.load(path2+'/'+path).get_fdata() for path in controls])
oautism_scans = np.array([nib.load(path1+'/'+path).get_fdata() for path in autism])
#Output variable
y=np.zeros(800)
y[400:800]=1
y.shape
#Multiple input variables or features
Xo=np.concatenate((oautism_scans[0:400], ocontrol_scans[0:400]))
Xo_train, Xo_test, yo_train, yo_test = train_test_split(Xo, y, test_size=0.25, random_state=1)
Xo_train, Xo_val, yo_train, yo_val = train_test_split(Xo_train, yo_train, test_size=(1/3), random_state=1) # 0.25 x 0.8 = 0.2
#Base LeNet model for neural networks
lenet5 = keras.models.Sequential()
lenet5.add(keras.layers.Conv2D(filters=32, kernel_size=(5,5), padding='same', activation='relu', input_shape=(61, 73, 61)))
lenet5.add(keras.layers.MaxPool2D(strides=2))
lenet5.add(keras.layers.Conv2D(filters=48, kernel_size=(5,5), padding='valid', activation='relu'))
lenet5.add(keras.layers.MaxPool2D(strides=2))
lenet5.add(keras.layers.Flatten())
lenet5.add(keras.layers.Dense(256, activation='relu'))
lenet5.add(keras.layers.Dense(84, activation='relu'))
lenet5.add(keras.layers.Dense(10, activation='softmax'))
lenet5.compile(loss="sparse_categorical_crossentropy",
optimizer="sgd",
metrics=['accuracy'])
#Base LeNet model for neural networks -3D version
mlenet5 = keras.models.Sequential()
mlenet5.add(keras.layers.Conv3D(filters=32, kernel_size=(5,5,5), padding='same', activation='relu', input_shape=(61, 73, 61,1)))
mlenet5.add(keras.layers.MaxPool3D(strides=2))
mlenet5.add(keras.layers.Conv3D(filters=48, kernel_size=(5,5,5), padding='valid', activation='relu'))
mlenet5.add(keras.layers.MaxPool3D(strides=2))
mlenet5.add(keras.layers.Flatten())
mlenet5.add(keras.layers.Dense(256, activation='relu'))
mlenet5.add(keras.layers.Dense(84, activation='relu'))
mlenet5.add(keras.layers.Dense(10, activation='softmax'))
mlenet5.compile(loss="sparse_categorical_crossentropy",
optimizer="sgd",
metrics=['accuracy'])
#Base LeNet model for neural networks -3D version with size 10 kernel, double filter size and added layer
m2lenet5 = keras.models.Sequential()
m2lenet5.add(keras.layers.Conv3D(filters=64, kernel_size=(10,10,10), padding='same', activation='relu', input_shape=(61, 73, 61,1)))
m2lenet5.add(keras.layers.MaxPool3D(strides=2))
m2lenet5.add(keras.layers.Conv3D(filters=96, kernel_size=(10,10,10), padding='valid', activation='relu'))
m2lenet5.add(keras.layers.MaxPool3D(strides=2))
m2lenet5.add(keras.layers.Flatten())
#added layer
m2lenet5.add(keras.layers.Dense(1024, activation='relu'))
m2lenet5.add(keras.layers.Dense(256, activation='relu'))
m2lenet5.add(keras.layers.Dense(84, activation='relu'))
m2lenet5.add(keras.layers.Dense(10, activation='softmax'))
m2lenet5.compile(loss="sparse_categorical_crossentropy",
optimizer="sgd",
metrics=['accuracy'])
2022-06-01 17:29:40.744545: I tensorflow/core/platform/cpu_feature_guard.cc:193] This TensorFlow binary is optimized with oneAPI Deep Neural Network Library (oneDNN) to use the following CPU instructions in performance-critical operations: AVX2 FMA To enable them in other operations, rebuild TensorFlow with the appropriate compiler flags.
lenet5.summary()
Model: "sequential"
_________________________________________________________________
Layer (type) Output Shape Param #
=================================================================
conv2d (Conv2D) (None, 61, 73, 32) 48832
max_pooling2d (MaxPooling2D (None, 30, 36, 32) 0
)
conv2d_1 (Conv2D) (None, 26, 32, 48) 38448
max_pooling2d_1 (MaxPooling (None, 13, 16, 48) 0
2D)
flatten (Flatten) (None, 9984) 0
dense (Dense) (None, 256) 2556160
dense_1 (Dense) (None, 84) 21588
dense_2 (Dense) (None, 10) 850
=================================================================
Total params: 2,665,878
Trainable params: 2,665,878
Non-trainable params: 0
_________________________________________________________________
mlenet5.summary()
Model: "sequential_1"
_________________________________________________________________
Layer (type) Output Shape Param #
=================================================================
conv3d (Conv3D) (None, 61, 73, 61, 32) 4032
max_pooling3d (MaxPooling3D (None, 30, 36, 30, 32) 0
)
conv3d_1 (Conv3D) (None, 26, 32, 26, 48) 192048
max_pooling3d_1 (MaxPooling (None, 13, 16, 13, 48) 0
3D)
flatten_1 (Flatten) (None, 129792) 0
dense_3 (Dense) (None, 256) 33227008
dense_4 (Dense) (None, 84) 21588
dense_5 (Dense) (None, 10) 850
=================================================================
Total params: 33,445,526
Trainable params: 33,445,526
Non-trainable params: 0
_________________________________________________________________
m2lenet5.summary()
Model: "sequential_2"
_________________________________________________________________
Layer (type) Output Shape Param #
=================================================================
conv3d_2 (Conv3D) (None, 61, 73, 61, 64) 64064
max_pooling3d_2 (MaxPooling (None, 30, 36, 30, 64) 0
3D)
conv3d_3 (Conv3D) (None, 21, 27, 21, 96) 6144096
max_pooling3d_3 (MaxPooling (None, 10, 13, 10, 96) 0
3D)
flatten_2 (Flatten) (None, 124800) 0
dense_6 (Dense) (None, 1024) 127796224
dense_7 (Dense) (None, 256) 262400
dense_8 (Dense) (None, 84) 21588
dense_9 (Dense) (None, 10) 850
=================================================================
Total params: 134,289,222
Trainable params: 134,289,222
Non-trainable params: 0
_________________________________________________________________
lenet5model = lenet5.fit(Xo_train, yo_train, validation_data=(Xo_val, yo_val))
13/13 [==============================] - 5s 335ms/step - loss: 1.8623 - accuracy: 0.4200 - val_loss: 0.9292 - val_accuracy: 0.4950
lenet5model3d = mlenet5.fit(Xo_train, yo_train, validation_data=(Xo_val, yo_val))
13/13 [==============================] - 553s 42s/step - loss: 1.4652 - accuracy: 0.4800 - val_loss: 1.0074 - val_accuracy: 0.4950
lenet5deepmodel3d = m2lenet5.fit(Xo_train, yo_train, validation_data=(Xo_val, yo_val))
13/13 [==============================] - 3767s 288s/step - loss: 1.8900 - accuracy: 0.4875 - val_loss: 0.8922 - val_accuracy: 0.5050
lenet5.evaluate(Xo_test, yo_test)
7/7 [==============================] - 0s 48ms/step - loss: 0.9209 - accuracy: 0.4700
[0.9208614230155945, 0.4699999988079071]
mlenet5.evaluate(Xo_test, yo_test)
7/7 [==============================] - 20s 3s/step - loss: 1.0642 - accuracy: 0.4650
[1.0641756057739258, 0.4650000035762787]
m2lenet5.evaluate(Xo_test, yo_test)
7/7 [==============================] - 147s 21s/step - loss: 0.8377 - accuracy: 0.5350
[0.8376597762107849, 0.5350000262260437]
import visualkeras
visualkeras.layered_view(lenet5)
visualkeras.layered_view(m2lenet5)
lenet5.get_weights()[5].shape
(256,)
mlenet5.get_weights()[5].shape
(256,)
m2lenet5.get_weights()[5].shape
(1024,)
lenet5.layers[7].output
<KerasTensor: shape=(None, 10) dtype=float32 (created by layer 'dense_2')>
#Creating embedding representations with the flat layer of 256 paramters for the base LeNet5 model
from keras import backend
lenet5_embed_model = backend.function([lenet5.layers[0].input],
[lenet5.layers[5].output])
aut_lenet5_embed =lenet5_embed_model([oautism_scans])[0]
cont_lenet5_embed =lenet5_embed_model([ocontrol_scans])[0]
aut_lenet5_embed.shape
(408, 256)
oautism_scans.shape
(408, 61, 73, 61)
#representing the embedding representation using TSNE transformation
from sklearn.manifold import TSNE
tsne = TSNE(n_components=3, random_state=0)
autism_tsne = tsne.fit_transform(aut_lenet5_embed)
controls_tsne = tsne.fit_transform(cont_lenet5_embed)
/usr/local/lib/python3.9/site-packages/sklearn/manifold/_t_sne.py:780: FutureWarning: The default initialization in TSNE will change from 'random' to 'pca' in 1.2. warnings.warn( /usr/local/lib/python3.9/site-packages/sklearn/manifold/_t_sne.py:790: FutureWarning: The default learning rate in TSNE will change from 200.0 to 'auto' in 1.2. warnings.warn( /usr/local/lib/python3.9/site-packages/sklearn/manifold/_t_sne.py:780: FutureWarning: The default initialization in TSNE will change from 'random' to 'pca' in 1.2. warnings.warn( /usr/local/lib/python3.9/site-packages/sklearn/manifold/_t_sne.py:790: FutureWarning: The default learning rate in TSNE will change from 200.0 to 'auto' in 1.2. warnings.warn(
#Plotting the TSNE transformation
plt.figure(figsize=(50, 50))
fig = plt.figure(figsize = (10, 10))
ax = fig.add_subplot(projection='3d')
ax.scatter(autism_tsne[:,0],autism_tsne[:,1],autism_tsne[:,2], color='red')
ax.scatter(controls_tsne[:,1], controls_tsne[:,0], controls_tsne[:,2],color='blue')
plt.legend(["Autism","Control"], loc='right')
plt.show()
<Figure size 3600x3600 with 0 Axes>
#Creating a 1D reprentation for the input image
X_train=Xo_train.reshape(400,-1)
X_val=Xo_val.reshape(200,-1)
X_model=np.concatenate((X_train, X_val))
y_model=np.concatenate((yo_train, yo_val))
X_test=Xo_test.reshape(200,-1)
#Random Forest
cmodel = RandomForestClassifier().fit(X_model, y_model)
cscoresr2 = cross_validate(cmodel, X_model, y_model, scoring="accuracy", cv=crossvalidation, n_jobs=10, return_train_score=True)
print("Random Forest Training "+"Folds: " + str(len(cscoresr2['train_score'])) + ", Accuracy: " + str(np.mean(np.abs(cscoresr2['train_score']))) + ", STD: " + str(np.std(cscoresr2['train_score'])))
print("Random Forest Test "+"Folds: " + str(len(cscoresr2['test_score'])) + ", Accuracy: " + str(np.mean(np.abs(cscoresr2['test_score']))) + ", STD: " + str(np.std(cscoresr2['test_score'])))
Random Forest Training Folds: 5, AUC: 1.0, STD: 0.0 Random Forest Test Folds: 5, AUC: 0.5166666666666667, STD: 0.026352313834736484
from sklearn.metrics import accuracy_score, classification_report
y_randfor = cmodel.predict(X_test)
y_neural = m2lenet5.predict(X_test)
print("Acccuracy score with Random Forest: "+str(accuracy_score(yo_test, y_pred_test)))
Acccuracy score with Random Forest: 0.565
confusion_matrix(yo_test, y_randfor)
array([[59, 34],
[53, 54]])
print(classification_report(yo_test, y_randfor))
precision recall f1-score support
0.0 0.53 0.63 0.58 93
1.0 0.61 0.50 0.55 107
accuracy 0.56 200
macro avg 0.57 0.57 0.56 200
weighted avg 0.57 0.56 0.56 200
y_neural = m2lenet5.predict(Xo_test)
7/7 [==============================] - 148s 21s/step
confusion_matrix(yo_test, y_neural_bool)
array([[ 0, 93],
[ 0, 107]])
y_neural_bool=np.argmax(y_neural,axis=1)
print(classification_report(yo_test, y_neural_bool))
precision recall f1-score support
0.0 0.00 0.00 0.00 93
1.0 0.54 1.00 0.70 107
accuracy 0.54 200
macro avg 0.27 0.50 0.35 200
weighted avg 0.29 0.54 0.37 200
/usr/local/lib/python3.9/site-packages/sklearn/metrics/_classification.py:1318: UndefinedMetricWarning: Precision and F-score are ill-defined and being set to 0.0 in labels with no predicted samples. Use `zero_division` parameter to control this behavior. _warn_prf(average, modifier, msg_start, len(result)) /usr/local/lib/python3.9/site-packages/sklearn/metrics/_classification.py:1318: UndefinedMetricWarning: Precision and F-score are ill-defined and being set to 0.0 in labels with no predicted samples. Use `zero_division` parameter to control this behavior. _warn_prf(average, modifier, msg_start, len(result)) /usr/local/lib/python3.9/site-packages/sklearn/metrics/_classification.py:1318: UndefinedMetricWarning: Precision and F-score are ill-defined and being set to 0.0 in labels with no predicted samples. Use `zero_division` parameter to control this behavior. _warn_prf(average, modifier, msg_start, len(result))