Computer Vision Based Skin Cancer Classification by Using Texture Features

.


Introduction
The growth and division of healthy cells result in the formation of new cells.These new cells will eventually take the place of injured or elderly cells.Cancer is the name of the disease that occurs when the body's normal cells multiply uncontrollably.Cancer cells, rather than killing off, continue to proliferate and give rise to more aberrant cells.A tumor can develop when there is aberrant cell proliferation in the epidermal layer, which may destroy other healthy tissues.This process, known as metastasis, describes how cancer cells move from one location in the body to another through the circulatory and lymphatic systems.Cancer comes in a wide variety of forms [1].
The most frequent types of cancer include thyroid cancer, lung cancer, breast cancer, bladder cancer, kidney cancer, colon and rectum cancer, leukemia, pancreatic cancer, prostate cancer, and skin cancer.Other less common types of cancer include leukemia and pancreatic cancer.The surface of our skin is the most significant organ in our bodies [2].This organ's primary responsibilities include acting as a protective barrier against potentially hazardous substances that may enter the body from the environment, regulating body temperature via the hair, sweat glands, and adipose tissue that it contains, ensuring proper fluid and electrolyte balance, and contributing to the overall health of our bodies.The layers of our skin may be broken down into three broad categories.The epidermis, the topmost layer of our skin, is the first layer we will discuss [3].
In most cases, this layer stops the loss of fluid and serves as protection for the tissues that lie underneath it.The dermis is the second layer of the skin.In addition to hair follicles and sebaceous glands, the dermis is home to blood vessels and nerve fibers.It is in charge of regulating the https://mol2net-08.sciforum.net/temperature and maintaining the fluid-electrolyte balance [4].The hypodermis is the third and last layer of our skin.This layer is composed of adipose tissue.It not only guards against the damaging effects of impact on structures like bone and muscle but also helps maintain a steady body temperature.These three layers of our skin are fertile ground for the growth of tumors.The specific sort of cancer that an individual has depends on which tissues and layers of their skin the altered cells originate in.As a result, the procedure of treating the kind of cancer that was detected in the patient may be challenging or straightforward [5].
Skin cancer is the most prevalent form of the disease in every region of the globe surveyed by the Skin Cancer Foundation (SCF).According to the statistics from the Skin Cancer Foundation (SCF), more than two persons lose their lives to skin cancer every hour in the United States [6].Feature extraction is commonly used to classify segmented pictures in the training and test sets.This research extracted features using texture analysis [8].Asymmetry is a crucial indicator of skin lesion malignancy.It measures how comparable the lesion form is along the primary axis.

Materials and Methods
Vertical and horizontal asymmetry are computed independently.The compact index (CI) measures border irregularity as the ratio of the lesion circumference square to its area.Color is crucial to skin disease diagnosis.The normalized standard deviation of lesion red, green, and blue components measures color variance.The texture rule defines diameter as the most significant distance between any two sites of the lesion boundary.The lesion diameter is the diameter of the circle with the lesion area.Image texture determines pattern and color constancy.Haralick texture characterizes texturebased images by computing the gray-level co-occurrence matrix (GLCM).GLCM is used to extract texture features since it can calculate many characteristics and is easy to use.The co-occurrence matrix extracts 14 characteristic texture properties from the probability matrix.This research selected https://mol2net-08.sciforum.net/Haralick's contrast, correlation, energy, and homogeneity [9].GLCM is based on picture-pixel neighborhoods.It records the complete picture by searching for adjacent pairs of pixel values.
Correlation measures the combined probability of pixel pairings in each row and column.
Energy, the square root of the total of square pixels, is the recurrence of pairs of pixels in the picture.
Contrast distinguishes items by hue or color.Homogeneity; measures how near the GLCM component distribution is to its diagonal [10].

Results and Discussion
.
The ISIC 2020 training set, HAM10000, identified seven skin disorders in this investigation[7].The dataset contains 300 150x2 RGB skin lesion photos.The significant number of lesions per category separates this dataset from others.These photos were classed as abnormal and usual.The dataset contains 300 malignant and 300 benign skin lesions.Preprocessing Ham10000 photos improved picture quality.Images received contrast-limited adaptive histogram equalization, morphological occlusion, and median filter.Image segmentation follows preprocessing.Image segmentation enhances meaning and analysis.Segmentation typically determines analysis success.Medical picture segmentation requires ROI extraction.Area-based segmentation uses the watershed transform to define the region of interest (ROI) and choose the ROI closest to the skin lesion.


Support vector machine (SVM) Time taken to build the model: 0.38 seconds  mode: 10 fold

Figure 1 :
Figure 1: Accuracy of Dataset using SVM Classifier

Table 2 :
SVM Classifier Detailed Accuracy

Table 3 :
Confusion Matrix result using SVM Classifier