A Review of Recent Machine Learning Techniques Used for Skin Lesion Image Classification

A Review of Recent Machine Learning Techniques Used for Skin Lesion Image Classification

Mayank Upadhyay, Jyoti Rawat, Kriti
Copyright: © 2023 |Pages: 15
DOI: 10.4018/978-1-6684-6957-6.ch005
OnDemand:
(Individual Chapters)
Available
$37.50
Current Special Offers
No Current Special Offers
TOTAL SAVINGS: $37.50

Abstract

Skin cancer is amongst the most common forms of cancer and can become life-threatening if not detected early. Due to the rise in the number of cancer cases, there is a growing interest in using computational diagnostics for early cancer detection as the specificity rate of even an expert dermatologist is around 59%. Computer-aided diagnosis can significantly contribute to skin lesion image analysis. Skin cancer prognostication can be achieved with a classification that assigns data objects to particular classes based on extracted features. The steps for image classification are pre-processing where noise is removed and lesion features are highlighted, making it easier to classify the image, detection of the lesion on skin (i.e. segmentation), extracting useful features, and finally applying classification algorithm. This paper provides a review of the recent studies in the bailiwick of skin cancer image classification using machine learning (ML) algorithms.
Chapter Preview
Top

Background

ML techniques can play a pivotal role in the public healthcare sector, as they can give high accuracy in the prognosis of skin cancer. Typically, the steps involved in lesion image classification are as shown in figure 1. This paper presents a review of recent research performed in the area of skin lesion classification using ML algorithms.

Figure 1.

Steps involved in classifying the skin lesion image dataset

978-1-6684-6957-6.ch005.f01

Data Acquisition

Before the pre-processing step, the image dataset needs to be collected. There are many skin cancer image repositories available online which are widely used in cancer image classification research. The dataset used for training the model should be sufficiently large (Petrellis, 2018) and if the dataset is imbalanced then techniques like data augmentation (Vidya & Karki, 2020) and SMOTE (Chakravorty et al., 2016) can be used. However, one of the limitations in the reviewed literature was the small size of the dataset used. Figure 2 shows the size of the dataset used in the reviewed literature, as 59% of the literature used dataset containing no. of images only between 1 to 200.

Complete Chapter List

Search this Book:
Reset