New approach to identify some skin diseases

We present two methods used in machine learning, the deep convolution neural networks, where input is the images of skin diseases. This helps us in encoding certain characteristics in the image and thus training and classification. and the Particle Swarm Optimization algorithm for the optimal solution to improve the structure of the neural network and reduce the time to discover the disease.

Deep Convolution Neural Network (CNN) [2, C. 303]:

A Convolutional Neural Network (CNN) is the foundation of most computer vision technologies. Unlike traditional multilayer perceptron architectures, it uses two operations called `convolution' and pooling' to reduce an image into its essential features, and uses those features to understand and classify the image.

The basic building blocks of CNN are:

- Convolution layer: a “filter”, sometimes called a “kernel”, is passed over the image, viewing a few pixels at a time (for example, 3X3 or 5X5). The convolution operation is a dot product of the original pixel values with weights defined in the filter. The results are summed up into one number that represents all the pixels the filter observed.

- Activation layer: the convolution layer generates a matrix that is much smaller in size than the original image. This matrix is run through an activation layer, which introduces non-linearity to allow the network to train itself via backpropagation. The activation function is typically ReLu.

- Pooling layer: “pooling” is the process of further down sampling and reducing the size of the matrix. A filter is passed over the results of the previous layer and selects one number out of each group of values (typically the maximum, this is called max pooling). This allows the network to train much faster, focusing on the most important information in each feature of the image.

- Fully connected layer: a traditional multilayer perceptron structure. Its input is a onedimensional vector representing the output of the previous layers. Its output is a list of probabilities for different possible labels attached to the image.The label that receives the highest probability is the classification decision.

There may be multiple activation and pooling layers, depending on the CNN architecture

Figure (1) define CNN architecture

Particle Swarm Optimization Algorithm (PSO) [3, C. 55]:

Particle Swarm Optimization is an evolutionary computation technique originally developed by Kennedy and Eberhart (1995). The PSO is motivated from the simulation of social behavior instead of evolution of nature as in the other evolutionary algorithms (genetic algorithms, evolutionary programming, evolutionary strategies, and genetic programming). PSO is sociologically inspired, since the algorithm is based on sociological behavior associated with bird flocking. It is a population based evolutionary algorithm. Similar to the other population based evolutionary algorithms.

PSO algorithm consists of the following steps:

Step 1: Initialization: Initialize a population of particles with random position and velocities in d dimensional problem space. Confine the search space by specifying the lower and upper limits of each decision variable. The populations of points are initialized with the velocity and position set to fall into the pre-specified or allowed range and satisfying the equality and inequality constraints.

Step 2: Velocity updating: At each iteration, the velocities of all particles are updated according to the equation (1) which is:

equation (1)

where V^(t)id and X^(t)id are the velocity and position of particle i, in d dimensional space respectively. P^(t)bestid best position of individual i in dimensional space until generation t; g^(t)best d is the best position of the group in d dimension until generation t; w is the inertia weight factor controlling the dynamics of flying; cl and c2 are accelerating constants; randl and rand2 are random variables in the range [0,1].

The first part of the equation (1) is the momentum part of the particle. The inertia weight w represents the degree of the momentum of the particles. The second part of the equation (1) is the cognition part which represents the independent thinking of the particle itself. The third part of equation (1) is the social part which represents the collaboration among the particles.

Figure (2) concept of changing particle position in PSO

Step 3: Position updating: Between successive iterations, the position of all particles are updated according to the equation (2) which is

where id^(t+1)x is the new position and X^(t)id is the previous position and V^(t+1)id is the new velocity.

Step 4: Memory updating: Update particle best position P^(t)best id and global best position g^best d using equation (3).

where f(X) is the objective function to be minimized. Compare particles fitness evaluation with particles p^(t)best id. If current value is better than p^(t)best id then set p^(t+¹⁾best id value equal to the current value and the p^(t+¹⁾best id location equal to the current location in d dimensional space. Compare fitness evaluation with the population's overall previous best. If the current value is better than g best d then reset g^(t+¹⁾best d to the current particles array index and value.

Step 5: Termination criteria examination: The algorithm repeats Step 2 to Step 4 until a sufficient good fitness or a maximum number of iterations/epochs are reached. Once terminated, the algorithm outputs the points of g^(t)best d and f(g^(t)best d ) as its solution.

Practical application

Identifying skin diseases:

Skin cancer is one of the most common forms of cancer affecting humans and early diagnosis is extremely vital in curing the disease [1, C. 49]. So far, the human knowledge in this field is very limited, thus, developing a mechanism capable of identifying the disease early on can save lives, reduce intervention and cut unnecessary costs, so we developed a system to classify skin caner.

Skin cancer can be of three types:

• Basal cell carcinoma

• Squamous cell carcinoma

• Melanoma cancer

Collecting pictures of diseases:

Pictures of skin diseases were collected under the supervision of a specialist, 3000 color pictures as follows:

• Basal cell carcinoma: 1,000 images

• Squamous cell carcinoma: 1,000 images

• Melanoma cancer: 1,000 images

Image Processing:

Usually skin pictures or digital photos taken from regular camera contain noises like hair, air bubbles etc. This noise may result in inaccurate classification and will give the system a false prediction result. In order to avoid this, the images are subjected to various image processing techniques. We applied gaussian filter to remove the noise as in Figure(3).

Figure(3) melanoma cancer after applying the filter

Deep Convolution Neural Network:

Using the 80/20 rule for training and testing, for each disease, 800 pictures are selected for each disease in the training phase and 200 pictures in the testing phase.

Table

Specifications of the device on which the network was trained and tested

The arrangement of the network layers is as follows: After training the network, the results were as follows:

Using the Particle Swarm Optimization algorithm to find the optimal structure of Convolution Neural Network:

define the lower bounds and upper bounds of each layer parameters which have been optimized to get the optimal CNN structure:

Filter size = [3 - 5]

Filter No = [8 - 32]

Neurons No = [1024 - 2048]

Population size =1000

After apply Particle Swarm Optimization algorithm depending on above layer parameters, we get the optimal solution:

After get the best solution,apply this on CNN and we get the result:

Ш Accuracy: 100%

Ш Training time: 50 minutes 2 second

We notice that when applying the Particle Swarm Optimization algorithm, the time decreased to about 20 minutes.

Compare Results

There are different research works which have been introduced for skin cancer detection. Each of these methods have their own difficulties and shortcomings. Introducing all of these methods is not possible. Therefore 4 methods have been selected for comparison with our proposed method. Some deep learning based systems like AlexNet[4, C. 1101],VGG-16[5, C. 1410], ResNet[6, C. 95], and Inception-v3[7, C]

Table 1

Comparison of the accuracy for skin cancer detection

Table 3

illustrates a accuracy comparison between the proposed system and the aforesaid methods

As can be observed, the CNN/PSO method is most accurate when compared with the other 4 aforesaid methods.

Conclusions