Skin Cancer is one of the most common types of disease in the United States. Up to 4 Million cases have been reported dead due to skin cancer in the United States over the year. In this article, I will create a model for skin cancer classification with Machine Learning.
This is a huge number, really 4 million, people died just from Skin cancer in a country. As all those people have been dying, but half of those cases or maybe even more, didn’t have went to the doctor at the early stages of the disease when it might have been prevented.
If people are getting the symptoms of skin cancer, they still don’t reach out for a doctor, which is not a good signal because skin cancer can be cured at the early stages.
Also, read – Naive Bayes Classification with Machine Learning
Skin Cancer Classification with Machine Learning
So this is where a machine learning algorithm works in the classification of Skin Cancer. As I mentioned earlier Skin Cancer can be easily cured in the early stages of the disease, but it’s the people who don’t want to visit the doctor.
So here is a simple machine learning algorithm, which can help those people to identify if they are having skin cancer or not while sitting at their homes. This Machine Learning algorithm is based on Convolutional Neural Networks (CNN).
CNN Layers Classification for Skin Cancer Detection
Lets start with importing the libraries
import numpy as np
from skimage import io
import matplotlib.pyplot as pltCode language: Python (python)Now I will simply upload images to train our machine learning model using the skimage library in python.
imgb = io.imread('bimg-1049.png')
imgm = io.imread('mimg-178.png')
imgb = io.imread('bimg-721.png')
imgm = io.imread('mimg-57.png')Code language: Python (python)You can download these images from below
These images are sample images of Benign mole and Malign mole which are a type of skin problems.
Let’s visualize these images
plt.figure(figsize=(10,20))
plt.subplot(121)
plt.imshow(imgb)
plt.axis('off')
plt.subplot(122)
plt.imshow(imgm)
plt.axis('off')Code language: Python (python)
Now lets train our model for further classification
from keras.models import load_model
model = load_model('BM_VA_VGG_FULL_DA.hdf5')
from keras import backend as K
def activ_viewer(model, layer_name, im_put):
layer_dict = dict([(layer.name, layer) for layer in model.layers])
layer = layer_dict[layer_name]
activ1 = K.function([model.layers[0].input, K.learning_phase()], [layer.output,])
activations = activ1((im_put, False))
return activations
def normalize(x):
# utility function to normalize a tensor by its L2 norm
return x / (K.sqrt(K.mean(K.square(x))) + 1e-5)
def deprocess_image(x):
# normalize tensor: center on 0., ensure std is 0.1
x -= x.mean()
x /= (x.std() + 1e-5)
x *= 0.1
# clip to [0, 1]
x += 0.5
x = np.clip(x, 0, 1)
# convert to RGB array
x *= 255
if K.image_data_format() == 'channels_first':
x = x.transpose((1, 2, 0))
x = np.clip(x, 0, 255).astype('uint8')
return x
def plot_filters(filters):
newimage = np.zeros((16*filters.shape[0],8*filters.shape[1]))
for i in range(filters.shape[2]):
y = i%8
x = i//8
newimage[x*filters.shape[0]:x*filters.shape[0]+filters.shape[0],
y*filters.shape[1]:y*filters.shape[1]+filters.shape[1]] = filters[:,:,i]
plt.figure(figsize = (10,20))
plt.imshow(newimage)
plt.axis('off')Code language: Python (python)layer_dict = dict([(layer.name, layer) for layer in model.layers])Code language: Python (python)Skin Cancer Classification Model Summary
To see the summary of our trained model we will execute the code below
model.summary()Code language: Python (python)_________________________________________________________________ Layer (type) Output Shape Param # ================================================================= input_1 (InputLayer) (None, 128, 128, 3) 0 _________________________________________________________________ block1_conv1 (Conv2D) (None, 128, 128, 64) 1792 _________________________________________________________________ block1_conv2 (Conv2D) (None, 128, 128, 64) 36928 _________________________________________________________________ block1_pool (MaxPooling2D) (None, 64, 64, 64) 0 _________________________________________________________________ block2_conv1 (Conv2D) (None, 64, 64, 128) 73856 _________________________________________________________________ block2_conv2 (Conv2D) (None, 64, 64, 128) 147584 _________________________________________________________________ block2_pool (MaxPooling2D) (None, 32, 32, 128) 0 _________________________________________________________________ block3_conv1 (Conv2D) (None, 32, 32, 256) 295168 _________________________________________________________________ block3_conv2 (Conv2D) (None, 32, 32, 256) 590080 _________________________________________________________________ block3_conv3 (Conv2D) (None, 32, 32, 256) 590080 _________________________________________________________________ block3_pool (MaxPooling2D) (None, 16, 16, 256) 0 _________________________________________________________________ block4_conv1 (Conv2D) (None, 16, 16, 512) 1180160 _________________________________________________________________ block4_conv2 (Conv2D) (None, 16, 16, 512) 2359808 _________________________________________________________________ block4_conv3 (Conv2D) (None, 16, 16, 512) 2359808 _________________________________________________________________ block4_pool (MaxPooling2D) (None, 8, 8, 512) 0 _________________________________________________________________ block5_conv1 (Conv2D) (None, 8, 8, 512) 2359808 _________________________________________________________________ block5_conv2 (Conv2D) (None, 8, 8, 512) 2359808 _________________________________________________________________ block5_conv3 (Conv2D) (None, 8, 8, 512) 2359808 _________________________________________________________________ block5_pool (MaxPooling2D) (None, 4, 4, 512) 0 _________________________________________________________________ sequential_3 (Sequential) (None, 1) 2097665 ================================================================= Total params: 16,812,353 Trainable params: 16,812,353 Non-trainable params: 0
Now lets visualize the output of our trained model
activ_benign = activ_viewer(model,'block2_conv1',imgb.reshape(1,128,128,3))
img_benign = deprocess_image(activ_benign[0])
plot_filters(img_benign[0])
plt.figure(figsize=(20,20))
for f in range(128):
plt.subplot(8,16,f+1)
plt.imshow(img_benign[0,:,:,f])
plt.axis('off')Code language: Python (python)
activ_malign = activ_viewer(model,'block2_conv1',imgm.reshape(1,128,128,3))
img_malign = deprocess_image(activ_malign[0])
plot_filters(img_malign[0])
plt.figure(figsize=(20,20))
for f in range(128):
plt.subplot(8,16,f+1)
plt.imshow(img_malign[0,:,:,f])
plt.axis('off')Code language: Python (python)
Now lets zoom and visualize some of the filters
plt.figure(figsize=(10,10))
plt.subplot(121)
plt.imshow(img_benign[0,:,:,49])
plt.axis('off')
plt.subplot(122)
plt.imshow(img_malign[0,:,:,49])
plt.axis('off')Code language: Python (python)
plt.figure(figsize=(10,10))
plt.subplot(121)
plt.imshow(img_benign[0,:,:,94])
plt.axis('off')
plt.subplot(122)
plt.imshow(img_malign[0,:,:,94])
plt.axis('off')Code language: Python (python)
Testing the Model
Now let’s visualize and test the actual ability of our trained model
def plot_filters32(filters):
newimage = np.zeros((16*filters.shape[0],16*filters.shape[1]))
for i in range(filters.shape[2]):
y = i%16
x = i//16
newimage[x*filters.shape[0]:x*filters.shape[0]+filters.shape[0],
y*filters.shape[1]:y*filters.shape[1]+filters.shape[1]] = filters[:,:,i]
plt.figure(figsize = (15,25))
plt.imshow(newimage)
activ_benign = activ_viewer(model,'block3_conv3',imgb.reshape(1,128,128,3))
img_benign = deprocess_image(activ_benign[0])
plot_filters32(img_benign[0])Code language: Python (python)
activ_malign = activ_viewer(model,'block3_conv3',imgm.reshape(1,128,128,3))
img_malign = deprocess_image(activ_malign[0])
plot_filters32(img_malign[0])Code language: Python (python)
