Project: Covid19 classifier based on X-rays lung image
Published:
This blog post is about a project that focuses on building a deep learning model to predict whether patients have pneumonia, Covid-19, or no illness based on X-ray lung scans. The project is compatible with Google Colaboratory and uses TensorFlow 2.8.2. The objective of the project is to create a Covid-19 classifier using deep learning. The step-by-step instructions on how to preprocess the data, build the model architecture, and train and evaluate the model are provided.
The post is compatible with Google Colaboratory and can be accessed through this link:
Project: Covid19 classifier based on X-rays lung image
- Developed by Armin Norouzi
Compatible with Google Colaboratory- Tensorflow 2.8.2
- Objective: Build a deep learning model to predict that patients who had either pneumonia, Covid-19, or no illness based on X-ray lung scans.
import tensorflow as tf
from tensorflow import keras
from tensorflow.keras.models import Sequential
from tensorflow.keras.preprocessing.image import ImageDataGenerator
from sklearn.metrics import classification_report
from sklearn.metrics import confusion_matrix
from tensorflow.keras.callbacks import EarlyStopping
from tensorflow.keras import layers
import matplotlib.pyplot as plt
import numpy
Geting data and being familiar with data
Get the data
import zipfile
# Download zip file of pizza_steak images
!wget https://gitlab.com/arminny/ml_course_datasets/-/raw/main/Covid19-dataset.zip
# Unzip the downloaded file
zip_ref = zipfile.ZipFile("Covid19-dataset.zip", "r")
zip_ref.extractall()
zip_ref.close()
--2022-09-21 16:28:24-- https://gitlab.com/arminny/ml_course_datasets/-/raw/main/Covid19-dataset.zip
Resolving gitlab.com (gitlab.com)... 172.65.251.78, 2606:4700:90:0:f22e:fbec:5bed:a9b9
Connecting to gitlab.com (gitlab.com)|172.65.251.78|:443... connected.
HTTP request sent, awaiting response... 200 OK
Length: 21847859 (21M) [application/octet-stream]
Saving to: ‘Covid19-dataset.zip’
Covid19-dataset.zip 100%[===================>] 20.83M 63.9MB/s in 0.3s
2022-09-21 16:28:25 (63.9 MB/s) - ‘Covid19-dataset.zip’ saved [21847859/21847859]
!ls Covid19-dataset
test train
We can see we’ve got a train and test folder.
Let’s see what’s inside one of them.
!ls Covid19-dataset/train/
Covid Normal Pneumonia
import os
# Walk through pizza_steak directory and list number of files
for dirpath, dirnames, filenames in os.walk("Covid19-dataset"):
print(f"There are {len(dirnames)} directories and {len(filenames)} images in '{dirpath}'.")
There are 2 directories and 1 images in 'Covid19-dataset'.
There are 3 directories and 1 images in 'Covid19-dataset/test'.
There are 0 directories and 26 images in 'Covid19-dataset/test/Covid'.
There are 0 directories and 20 images in 'Covid19-dataset/test/Normal'.
There are 0 directories and 20 images in 'Covid19-dataset/test/Pneumonia'.
There are 3 directories and 1 images in 'Covid19-dataset/train'.
There are 0 directories and 111 images in 'Covid19-dataset/train/Covid'.
There are 0 directories and 70 images in 'Covid19-dataset/train/Normal'.
There are 0 directories and 70 images in 'Covid19-dataset/train/Pneumonia'.
# Get the class names
import pathlib
import numpy as np
data_dir = pathlib.Path("Covid19-dataset/train/") # turn our training path into a Python path
class_names = np.array(sorted([item.name for item in data_dir.glob('*')])) # created a list of class_names from the subdirectories
class_names = class_names[1:]
print(class_names)
['Covid' 'Normal' 'Pneumonia']
# View an image
import matplotlib.pyplot as plt
import matplotlib.image as mpimg
import random
def view_random_image(target_dir, target_class):
# Setup target directory (we'll view images from here)
target_folder = target_dir+target_class
# Get a random image path
random_image = random.sample(os.listdir(target_folder), 1)
# Read in the image and plot it using matplotlib
img = mpimg.imread(target_folder + "/" + random_image[0])
plt.imshow(img)
plt.title(target_class)
plt.axis("off");
print(f"Image shape: {img.shape}") # show the shape of the image
return img
# View a random image from the training dataset
img = view_random_image(target_dir="Covid19-dataset/train/",
target_class="Normal")
Image shape: (256, 256, 4)
# View a random image from the training dataset
img = view_random_image(target_dir="Covid19-dataset/train/",
target_class="Covid")
Image shape: (256, 256, 4)
# View a random image from the training dataset
img = view_random_image(target_dir="Covid19-dataset/train/",
target_class="Pneumonia")
Image shape: (256, 256, 4)
# View the image shape
img.shape # returns (width, height, colour channels)
(256, 256, 4)
Prepare data
#Construct an ImageDataGenerator object:
DIRECTORY = "Covid19-dataset/train/"
CLASS_MODE = "categorical"
COLOR_MODE = "grayscale"
TARGET_SIZE = (256,256)
Looking at the image shape more closely, you’ll see it’s in the form (Width, Height, Colour Channels).
BATCH_SIZE = 32
training_data_generator = ImageDataGenerator(rescale= 1/255,
zoom_range = 0.1,
rotation_range = 25,
height_shift_range = 0.05,
width_shift_range = 0.05)
validation_data_generator = ImageDataGenerator(rescale= 1/255)
training_iterator = training_data_generator.flow_from_directory(directory= "Covid19-dataset/train/",
target_size=(256, 256),
color_mode='grayscale',
class_mode='categorical',
batch_size=BATCH_SIZE)
validation_iterator = validation_data_generator.flow_from_directory(directory= "Covid19-dataset/test/",
target_size=(256, 256),
color_mode='grayscale',
class_mode='categorical',
batch_size=BATCH_SIZE)
Found 251 images belonging to 3 classes.
Found 66 images belonging to 3 classes.
Model structure
from tensorflow.keras.layers import Dense, Flatten, Conv2D, MaxPool2D, Dropout
from tensorflow.keras.optimizers import Adam
from tensorflow.keras import Sequential
from tensorflow.keras import losses
def model_structure():
model = Sequential()
model.add(Conv2D(10, 3, padding='same', input_shape=(256,256, 1), name = "Conv_1", activation="relu"))
model.add(Conv2D(10, 3, padding='same', name = "Conv_2", activation="relu"))
model.add(MaxPool2D(2, name = "maxpool_1"))
model.add(Conv2D(10, 5, padding='same', name = "Conv_3", activation="relu"))
model.add(Conv2D(10, 5, padding='same', name = "Conv_4", activation="relu"))
model.add(MaxPool2D(2, name = "maxpool_2"))
model.add(Conv2D(15, 7, padding='same', name = "Conv_5", activation="relu"))
model.add(MaxPool2D(2, name = "maxpool_3"))
model.add(Flatten())
model.add(Dense(3, activation="softmax"))
model.compile(optimizer = Adam(learning_rate=0.0001),
loss = losses.CategoricalCrossentropy(),
metrics=["accuracy"])
model.summary()
return model
model = model_structure()
Model: "sequential_6"
_________________________________________________________________
Layer (type) Output Shape Param #
=================================================================
Conv_1 (Conv2D) (None, 256, 256, 10) 100
Conv_2 (Conv2D) (None, 256, 256, 10) 910
maxpool_1 (MaxPooling2D) (None, 128, 128, 10) 0
Conv_3 (Conv2D) (None, 128, 128, 10) 2510
Conv_4 (Conv2D) (None, 128, 128, 10) 2510
maxpool_2 (MaxPooling2D) (None, 64, 64, 10) 0
Conv_5 (Conv2D) (None, 64, 64, 15) 7365
maxpool_3 (MaxPooling2D) (None, 32, 32, 15) 0
flatten_5 (Flatten) (None, 15360) 0
dense_5 (Dense) (None, 3) 46083
=================================================================
Total params: 59,478
Trainable params: 59,478
Non-trainable params: 0
_________________________________________________________________
Training
history = model.fit(training_iterator,
epochs = 100,
batch_size = BATCH_SIZE,
verbose = 1,
validation_data= validation_iterator)
Epoch 1/100
8/8 [==============================] - 3s 306ms/step - loss: 1.0960 - accuracy: 0.3466 - val_loss: 1.0890 - val_accuracy: 0.3939
Epoch 2/100
8/8 [==============================] - 2s 265ms/step - loss: 1.0762 - accuracy: 0.4422 - val_loss: 1.0753 - val_accuracy: 0.3939
Epoch 3/100
8/8 [==============================] - 2s 264ms/step - loss: 1.0552 - accuracy: 0.4422 - val_loss: 1.0591 - val_accuracy: 0.3939
Epoch 4/100
8/8 [==============================] - 2s 265ms/step - loss: 1.0247 - accuracy: 0.4422 - val_loss: 1.0165 - val_accuracy: 0.3939
Epoch 5/100
8/8 [==============================] - 2s 263ms/step - loss: 0.9731 - accuracy: 0.4661 - val_loss: 0.9445 - val_accuracy: 0.6212
Epoch 6/100
8/8 [==============================] - 2s 262ms/step - loss: 0.8888 - accuracy: 0.6255 - val_loss: 0.8396 - val_accuracy: 0.7273
Epoch 7/100
8/8 [==============================] - 2s 262ms/step - loss: 0.7710 - accuracy: 0.7331 - val_loss: 0.7271 - val_accuracy: 0.7121
Epoch 8/100
8/8 [==============================] - 2s 259ms/step - loss: 0.6256 - accuracy: 0.8127 - val_loss: 0.6646 - val_accuracy: 0.6970
Epoch 9/100
8/8 [==============================] - 2s 266ms/step - loss: 0.5209 - accuracy: 0.8167 - val_loss: 0.5964 - val_accuracy: 0.7121
Epoch 10/100
8/8 [==============================] - 2s 263ms/step - loss: 0.4598 - accuracy: 0.8446 - val_loss: 0.6125 - val_accuracy: 0.6818
Epoch 11/100
8/8 [==============================] - 2s 276ms/step - loss: 0.4147 - accuracy: 0.8566 - val_loss: 0.5854 - val_accuracy: 0.7424
Epoch 12/100
8/8 [==============================] - 2s 271ms/step - loss: 0.3614 - accuracy: 0.8606 - val_loss: 0.6405 - val_accuracy: 0.7121
Epoch 13/100
8/8 [==============================] - 2s 266ms/step - loss: 0.3243 - accuracy: 0.8805 - val_loss: 0.5697 - val_accuracy: 0.7424
Epoch 14/100
8/8 [==============================] - 2s 264ms/step - loss: 0.3975 - accuracy: 0.8446 - val_loss: 0.5427 - val_accuracy: 0.7576
Epoch 15/100
8/8 [==============================] - 2s 269ms/step - loss: 0.3453 - accuracy: 0.8606 - val_loss: 0.6043 - val_accuracy: 0.7424
Epoch 16/100
8/8 [==============================] - 2s 264ms/step - loss: 0.3489 - accuracy: 0.8765 - val_loss: 0.5033 - val_accuracy: 0.7727
Epoch 17/100
8/8 [==============================] - 2s 265ms/step - loss: 0.3358 - accuracy: 0.8606 - val_loss: 0.7202 - val_accuracy: 0.7273
Epoch 18/100
8/8 [==============================] - 2s 273ms/step - loss: 0.3518 - accuracy: 0.8765 - val_loss: 0.5462 - val_accuracy: 0.7576
Epoch 19/100
8/8 [==============================] - 2s 266ms/step - loss: 0.3099 - accuracy: 0.8845 - val_loss: 0.4603 - val_accuracy: 0.8030
Epoch 20/100
8/8 [==============================] - 2s 265ms/step - loss: 0.2534 - accuracy: 0.9203 - val_loss: 0.4926 - val_accuracy: 0.7879
Epoch 21/100
8/8 [==============================] - 2s 264ms/step - loss: 0.2760 - accuracy: 0.9084 - val_loss: 0.4660 - val_accuracy: 0.7879
Epoch 22/100
8/8 [==============================] - 2s 267ms/step - loss: 0.2380 - accuracy: 0.9004 - val_loss: 0.4541 - val_accuracy: 0.8030
Epoch 23/100
8/8 [==============================] - 2s 267ms/step - loss: 0.2664 - accuracy: 0.8964 - val_loss: 0.4381 - val_accuracy: 0.8030
Epoch 24/100
8/8 [==============================] - 2s 266ms/step - loss: 0.2927 - accuracy: 0.8924 - val_loss: 0.4307 - val_accuracy: 0.8182
Epoch 25/100
8/8 [==============================] - 2s 271ms/step - loss: 0.3163 - accuracy: 0.8645 - val_loss: 0.4715 - val_accuracy: 0.7879
Epoch 26/100
8/8 [==============================] - 2s 263ms/step - loss: 0.2634 - accuracy: 0.8964 - val_loss: 0.4497 - val_accuracy: 0.8030
Epoch 27/100
8/8 [==============================] - 2s 267ms/step - loss: 0.3057 - accuracy: 0.8805 - val_loss: 0.4060 - val_accuracy: 0.8030
Epoch 28/100
8/8 [==============================] - 2s 263ms/step - loss: 0.2516 - accuracy: 0.9124 - val_loss: 0.4946 - val_accuracy: 0.7879
Epoch 29/100
8/8 [==============================] - 2s 264ms/step - loss: 0.2905 - accuracy: 0.8725 - val_loss: 0.3828 - val_accuracy: 0.8333
Epoch 30/100
8/8 [==============================] - 2s 266ms/step - loss: 0.2703 - accuracy: 0.8884 - val_loss: 0.4362 - val_accuracy: 0.8030
Epoch 31/100
8/8 [==============================] - 2s 266ms/step - loss: 0.2255 - accuracy: 0.9283 - val_loss: 0.3962 - val_accuracy: 0.8485
Epoch 32/100
8/8 [==============================] - 2s 268ms/step - loss: 0.2374 - accuracy: 0.9084 - val_loss: 0.3697 - val_accuracy: 0.8333
Epoch 33/100
8/8 [==============================] - 2s 270ms/step - loss: 0.2240 - accuracy: 0.9084 - val_loss: 0.4175 - val_accuracy: 0.8182
Epoch 34/100
8/8 [==============================] - 2s 266ms/step - loss: 0.2663 - accuracy: 0.9084 - val_loss: 0.3784 - val_accuracy: 0.8636
Epoch 35/100
8/8 [==============================] - 2s 269ms/step - loss: 0.2286 - accuracy: 0.9124 - val_loss: 0.4117 - val_accuracy: 0.8485
Epoch 36/100
8/8 [==============================] - 2s 260ms/step - loss: 0.2512 - accuracy: 0.9004 - val_loss: 0.4037 - val_accuracy: 0.8485
Epoch 37/100
8/8 [==============================] - 2s 263ms/step - loss: 0.2376 - accuracy: 0.8964 - val_loss: 0.3816 - val_accuracy: 0.8182
Epoch 38/100
8/8 [==============================] - 2s 259ms/step - loss: 0.2597 - accuracy: 0.9004 - val_loss: 0.3911 - val_accuracy: 0.8485
Epoch 39/100
8/8 [==============================] - 2s 264ms/step - loss: 0.2134 - accuracy: 0.9163 - val_loss: 0.3685 - val_accuracy: 0.8485
Epoch 40/100
8/8 [==============================] - 2s 265ms/step - loss: 0.2036 - accuracy: 0.9323 - val_loss: 0.3927 - val_accuracy: 0.8485
Epoch 41/100
8/8 [==============================] - 2s 263ms/step - loss: 0.2338 - accuracy: 0.9044 - val_loss: 0.3909 - val_accuracy: 0.8333
Epoch 42/100
8/8 [==============================] - 2s 263ms/step - loss: 0.1750 - accuracy: 0.9243 - val_loss: 0.3673 - val_accuracy: 0.8485
Epoch 43/100
8/8 [==============================] - 2s 260ms/step - loss: 0.1910 - accuracy: 0.9323 - val_loss: 0.3519 - val_accuracy: 0.8485
Epoch 44/100
8/8 [==============================] - 2s 261ms/step - loss: 0.1760 - accuracy: 0.9203 - val_loss: 0.3569 - val_accuracy: 0.8636
Epoch 45/100
8/8 [==============================] - 2s 262ms/step - loss: 0.1858 - accuracy: 0.9283 - val_loss: 0.3348 - val_accuracy: 0.8939
Epoch 46/100
8/8 [==============================] - 2s 268ms/step - loss: 0.2281 - accuracy: 0.9283 - val_loss: 0.3721 - val_accuracy: 0.8788
Epoch 47/100
8/8 [==============================] - 2s 260ms/step - loss: 0.1942 - accuracy: 0.9084 - val_loss: 0.3237 - val_accuracy: 0.8939
Epoch 48/100
8/8 [==============================] - 2s 261ms/step - loss: 0.2121 - accuracy: 0.9243 - val_loss: 0.3331 - val_accuracy: 0.8788
Epoch 49/100
8/8 [==============================] - 2s 259ms/step - loss: 0.1960 - accuracy: 0.9163 - val_loss: 0.3011 - val_accuracy: 0.8788
Epoch 50/100
8/8 [==============================] - 2s 257ms/step - loss: 0.2081 - accuracy: 0.9084 - val_loss: 0.2999 - val_accuracy: 0.9091
Epoch 51/100
8/8 [==============================] - 2s 269ms/step - loss: 0.2003 - accuracy: 0.9323 - val_loss: 0.3071 - val_accuracy: 0.8939
Epoch 52/100
8/8 [==============================] - 2s 258ms/step - loss: 0.2049 - accuracy: 0.9163 - val_loss: 0.3124 - val_accuracy: 0.8636
Epoch 53/100
8/8 [==============================] - 2s 260ms/step - loss: 0.1966 - accuracy: 0.9283 - val_loss: 0.3114 - val_accuracy: 0.8788
Epoch 54/100
8/8 [==============================] - 2s 260ms/step - loss: 0.1685 - accuracy: 0.9402 - val_loss: 0.2999 - val_accuracy: 0.8788
Epoch 55/100
8/8 [==============================] - 2s 258ms/step - loss: 0.1900 - accuracy: 0.9442 - val_loss: 0.3265 - val_accuracy: 0.8788
Epoch 56/100
8/8 [==============================] - 2s 267ms/step - loss: 0.1994 - accuracy: 0.9163 - val_loss: 0.2780 - val_accuracy: 0.8788
Epoch 57/100
8/8 [==============================] - 2s 259ms/step - loss: 0.1913 - accuracy: 0.9323 - val_loss: 0.2797 - val_accuracy: 0.8939
Epoch 58/100
8/8 [==============================] - 2s 264ms/step - loss: 0.2055 - accuracy: 0.9124 - val_loss: 0.3114 - val_accuracy: 0.8939
Epoch 59/100
8/8 [==============================] - 2s 260ms/step - loss: 0.1938 - accuracy: 0.9442 - val_loss: 0.2733 - val_accuracy: 0.8939
Epoch 60/100
8/8 [==============================] - 2s 266ms/step - loss: 0.1902 - accuracy: 0.9283 - val_loss: 0.2676 - val_accuracy: 0.8939
Epoch 61/100
8/8 [==============================] - 2s 269ms/step - loss: 0.1739 - accuracy: 0.9243 - val_loss: 0.2900 - val_accuracy: 0.9091
Epoch 62/100
8/8 [==============================] - 2s 264ms/step - loss: 0.1613 - accuracy: 0.9402 - val_loss: 0.2720 - val_accuracy: 0.9091
Epoch 63/100
8/8 [==============================] - 2s 263ms/step - loss: 0.2034 - accuracy: 0.9163 - val_loss: 0.3617 - val_accuracy: 0.8636
Epoch 64/100
8/8 [==============================] - 2s 266ms/step - loss: 0.1798 - accuracy: 0.9363 - val_loss: 0.3147 - val_accuracy: 0.8333
Epoch 65/100
8/8 [==============================] - 2s 263ms/step - loss: 0.2301 - accuracy: 0.9163 - val_loss: 0.2789 - val_accuracy: 0.8788
Epoch 66/100
8/8 [==============================] - 2s 266ms/step - loss: 0.2055 - accuracy: 0.9124 - val_loss: 0.2814 - val_accuracy: 0.8788
Epoch 67/100
8/8 [==============================] - 2s 263ms/step - loss: 0.1701 - accuracy: 0.9283 - val_loss: 0.2794 - val_accuracy: 0.8788
Epoch 68/100
8/8 [==============================] - 2s 266ms/step - loss: 0.1793 - accuracy: 0.9363 - val_loss: 0.3022 - val_accuracy: 0.8788
Epoch 69/100
8/8 [==============================] - 2s 262ms/step - loss: 0.1818 - accuracy: 0.9283 - val_loss: 0.2742 - val_accuracy: 0.8636
Epoch 70/100
8/8 [==============================] - 2s 269ms/step - loss: 0.1815 - accuracy: 0.9243 - val_loss: 0.2666 - val_accuracy: 0.8939
Epoch 71/100
8/8 [==============================] - 2s 261ms/step - loss: 0.1681 - accuracy: 0.9402 - val_loss: 0.3037 - val_accuracy: 0.8485
Epoch 72/100
8/8 [==============================] - 2s 262ms/step - loss: 0.1708 - accuracy: 0.9203 - val_loss: 0.2617 - val_accuracy: 0.8939
Epoch 73/100
8/8 [==============================] - 2s 262ms/step - loss: 0.2125 - accuracy: 0.9203 - val_loss: 0.3488 - val_accuracy: 0.8485
Epoch 74/100
8/8 [==============================] - 2s 261ms/step - loss: 0.1378 - accuracy: 0.9522 - val_loss: 0.2408 - val_accuracy: 0.9091
Epoch 75/100
8/8 [==============================] - 2s 264ms/step - loss: 0.1985 - accuracy: 0.9442 - val_loss: 0.2302 - val_accuracy: 0.8939
Epoch 76/100
8/8 [==============================] - 2s 261ms/step - loss: 0.1670 - accuracy: 0.9442 - val_loss: 0.2272 - val_accuracy: 0.8788
Epoch 77/100
8/8 [==============================] - 2s 259ms/step - loss: 0.1449 - accuracy: 0.9323 - val_loss: 0.2579 - val_accuracy: 0.8939
Epoch 78/100
8/8 [==============================] - 2s 260ms/step - loss: 0.2069 - accuracy: 0.9243 - val_loss: 0.3032 - val_accuracy: 0.8788
Epoch 79/100
8/8 [==============================] - 2s 267ms/step - loss: 0.2019 - accuracy: 0.8964 - val_loss: 0.2782 - val_accuracy: 0.9091
Epoch 80/100
8/8 [==============================] - 2s 263ms/step - loss: 0.1571 - accuracy: 0.9482 - val_loss: 0.2639 - val_accuracy: 0.8788
Epoch 81/100
8/8 [==============================] - 2s 263ms/step - loss: 0.1528 - accuracy: 0.9522 - val_loss: 0.2639 - val_accuracy: 0.8788
Epoch 82/100
8/8 [==============================] - 2s 260ms/step - loss: 0.1985 - accuracy: 0.9203 - val_loss: 0.2738 - val_accuracy: 0.8788
Epoch 83/100
8/8 [==============================] - 2s 265ms/step - loss: 0.1671 - accuracy: 0.9363 - val_loss: 0.2736 - val_accuracy: 0.8636
Epoch 84/100
8/8 [==============================] - 2s 265ms/step - loss: 0.1401 - accuracy: 0.9482 - val_loss: 0.2672 - val_accuracy: 0.8939
Epoch 85/100
8/8 [==============================] - 2s 262ms/step - loss: 0.1474 - accuracy: 0.9562 - val_loss: 0.2559 - val_accuracy: 0.8788
Epoch 86/100
8/8 [==============================] - 2s 264ms/step - loss: 0.1552 - accuracy: 0.9482 - val_loss: 0.2325 - val_accuracy: 0.8788
Epoch 87/100
8/8 [==============================] - 2s 266ms/step - loss: 0.1568 - accuracy: 0.9482 - val_loss: 0.2525 - val_accuracy: 0.8788
Epoch 88/100
8/8 [==============================] - 2s 263ms/step - loss: 0.1660 - accuracy: 0.9482 - val_loss: 0.2424 - val_accuracy: 0.8636
Epoch 89/100
8/8 [==============================] - 2s 264ms/step - loss: 0.1152 - accuracy: 0.9522 - val_loss: 0.2510 - val_accuracy: 0.8788
Epoch 90/100
8/8 [==============================] - 2s 264ms/step - loss: 0.1437 - accuracy: 0.9482 - val_loss: 0.2698 - val_accuracy: 0.8788
Epoch 91/100
8/8 [==============================] - 2s 265ms/step - loss: 0.1401 - accuracy: 0.9442 - val_loss: 0.2759 - val_accuracy: 0.8636
Epoch 92/100
8/8 [==============================] - 2s 261ms/step - loss: 0.1155 - accuracy: 0.9681 - val_loss: 0.2367 - val_accuracy: 0.8788
Epoch 93/100
8/8 [==============================] - 2s 262ms/step - loss: 0.1585 - accuracy: 0.9283 - val_loss: 0.3214 - val_accuracy: 0.8485
Epoch 94/100
8/8 [==============================] - 2s 267ms/step - loss: 0.2070 - accuracy: 0.9243 - val_loss: 0.3235 - val_accuracy: 0.8788
Epoch 95/100
8/8 [==============================] - 2s 255ms/step - loss: 0.1957 - accuracy: 0.9124 - val_loss: 0.4249 - val_accuracy: 0.8485
Epoch 96/100
8/8 [==============================] - 2s 262ms/step - loss: 0.1821 - accuracy: 0.9203 - val_loss: 0.3548 - val_accuracy: 0.8333
Epoch 97/100
8/8 [==============================] - 2s 255ms/step - loss: 0.1278 - accuracy: 0.9681 - val_loss: 0.3206 - val_accuracy: 0.8788
Epoch 98/100
8/8 [==============================] - 2s 265ms/step - loss: 0.1440 - accuracy: 0.9442 - val_loss: 0.3125 - val_accuracy: 0.8636
Epoch 99/100
8/8 [==============================] - 2s 267ms/step - loss: 0.1409 - accuracy: 0.9363 - val_loss: 0.3066 - val_accuracy: 0.8636
Epoch 100/100
8/8 [==============================] - 2s 256ms/step - loss: 0.1262 - accuracy: 0.9442 - val_loss: 0.2725 - val_accuracy: 0.8636
history.history.keys()
dict_keys(['loss', 'accuracy', 'val_loss', 'val_accuracy'])
import matplotlib.pyplot as plt
plt.figure(figsize=(10, 7))
plt.plot(history.history['accuracy'], label = "accuracy")
plt.plot(history.history['val_accuracy'], label= "val accuracy")
plt.xlabel("Epochs")
plt.ylabel("loss/accuracy")
plt.legend()
plt.show()
plt.figure(figsize=(10, 7))
plt.plot(history.history['loss'], label = "loss")
plt.plot(history.history['val_loss'], label= "val_loss")
plt.xlabel("Epochs")
plt.ylabel("loss/accuracy")
plt.legend()
plt.show()
model.evaluate(validation_iterator)
3/3 [==============================] - 0s 70ms/step - loss: 0.2725 - accuracy: 0.8636
[0.2724951207637787, 0.8636363744735718]
fine tunning model
def train_plot_model(model, Epochs):
history = model.fit(training_iterator,
epochs = Epochs,
batch_size = BATCH_SIZE,
verbose = 0,
validation_data= validation_iterator)
plt.figure(figsize=(10, 7))
plt.plot(history.history['accuracy'], label = "accuracy")
plt.plot(history.history['val_accuracy'], label= "val accuracy")
plt.xlabel("Epochs")
plt.ylabel("loss/accuracy")
plt.legend()
plt.show()
plt.figure(figsize=(10, 7))
plt.plot(history.history['loss'], label = "loss")
plt.plot(history.history['val_loss'], label= "val_loss")
plt.xlabel("Epochs")
plt.ylabel("loss/accuracy")
plt.legend()
plt.show()
return history, model
def model_structure_2():
model = Sequential()
model.add(Conv2D(10, 3, padding='same', input_shape=(256,256, 1), name = "Conv_1", activation="relu"))
model.add(Conv2D(10, 3, padding='same', name = "Conv_2", activation="relu"))
model.add(MaxPool2D(2, name = "maxpool_1"))
model.add(Conv2D(10, 5, padding='same', name = "Conv_3", activation="relu"))
model.add(Conv2D(10, 5, padding='same', name = "Conv_4", activation="relu"))
model.add(MaxPool2D(2, name = "maxpool_2"))
model.add(Conv2D(20, 7, padding='same', name = "Conv_5", activation="relu"))
model.add(Conv2D(20, 7, padding='same', name = "Conv_6", activation="relu"))
model.add(MaxPool2D(2, name = "maxpool_3"))
model.add(Conv2D(25, 7, padding='same', name = "Conv_7", activation="relu"))
model.add(MaxPool2D(2, name = "maxpool_4"))
model.add(Flatten())
model.add(Dense(3, activation="softmax"))
model.compile(optimizer = Adam(learning_rate=0.0001),
loss = losses.CategoricalCrossentropy(),
metrics=["accuracy"])
model.summary()
return model
model_2 = model_structure_2()
Model: "sequential_10"
_________________________________________________________________
Layer (type) Output Shape Param #
=================================================================
Conv_1 (Conv2D) (None, 256, 256, 10) 100
Conv_2 (Conv2D) (None, 256, 256, 10) 910
maxpool_1 (MaxPooling2D) (None, 128, 128, 10) 0
Conv_3 (Conv2D) (None, 128, 128, 10) 2510
Conv_4 (Conv2D) (None, 128, 128, 10) 2510
maxpool_2 (MaxPooling2D) (None, 64, 64, 10) 0
Conv_5 (Conv2D) (None, 64, 64, 20) 9820
Conv_6 (Conv2D) (None, 64, 64, 20) 19620
maxpool_3 (MaxPooling2D) (None, 32, 32, 20) 0
Conv_7 (Conv2D) (None, 32, 32, 25) 24525
maxpool_4 (MaxPooling2D) (None, 16, 16, 25) 0
flatten_9 (Flatten) (None, 6400) 0
dense_9 (Dense) (None, 3) 19203
=================================================================
Total params: 79,198
Trainable params: 79,198
Non-trainable params: 0
_________________________________________________________________
history_2, model_2 = train_plot_model(model_2, 200)
model_2.evaluate(validation_iterator)
3/3 [==============================] - 0s 54ms/step - loss: 0.2663 - accuracy: 0.9242
[0.2662919759750366, 0.9242424368858337]
Evaluation
def evaluation(model):
test_steps_per_epoch = numpy.math.ceil(validation_iterator.samples / validation_iterator.batch_size)
predictions = model.predict(validation_iterator, steps=test_steps_per_epoch)
test_steps_per_epoch = numpy.math.ceil(validation_iterator.samples / validation_iterator.batch_size)
predicted_classes = numpy.argmax(predictions, axis=1)
true_classes = validation_iterator.classes
class_labels = list(validation_iterator.class_indices.keys())
report = classification_report(true_classes, predicted_classes, target_names=class_labels)
print(report)
cm=confusion_matrix(true_classes,predicted_classes)
print(cm)
evaluation(model)
precision recall f1-score support
Covid 0.36 0.35 0.35 26
Normal 0.28 0.25 0.26 20
Pneumonia 0.22 0.25 0.23 20
accuracy 0.29 66
macro avg 0.29 0.28 0.28 66
weighted avg 0.29 0.29 0.29 66
[[ 9 6 11]
[ 8 5 7]
[ 8 7 5]]
evaluation(model_2)
precision recall f1-score support
Covid 0.50 0.50 0.50 26
Normal 0.42 0.40 0.41 20
Pneumonia 0.33 0.35 0.34 20
accuracy 0.42 66
macro avg 0.42 0.42 0.42 66
weighted avg 0.43 0.42 0.42 66
[[13 6 7]
[ 5 8 7]
[ 8 5 7]]