A Convolutional Neural Network Framework for Accurate Skin Cancer Detection

  • PDF / 1,947,857 Bytes
  • 21 Pages / 439.37 x 666.142 pts Page_size
  • 117 Downloads / 224 Views

DOWNLOAD

REPORT


A Convolutional Neural Network Framework for Accurate Skin Cancer Detection Karl Thurnhofer-Hemsi1,2

· Enrique Domínguez1,2

Accepted: 3 October 2020 © Springer Science+Business Media, LLC, part of Springer Nature 2020

Abstract Skin diseases have become a challenge in medical diagnosis due to visual similarities. Although melanoma is the best-known type of skin cancer, there are other pathologies that are the cause of many death in recent years. The lack of large datasets is one of the main difficulties to develop a reliable automatic classification system. This paper presents a deep learning framework for skin cancer detection. Transfer learning was applied to five stateof-art convolutional neural networks to create both a plain and a hierarchical (with 2 levels) classifiers that are capable to distinguish between seven types of moles. The HAM10000 dataset, a large collection of dermatoscopic images, were used for experiments, with the help of data augmentation techniques to improve performance. Results demonstrate that the DenseNet201 network is suitable for this task, achieving high classification accuracies and F-measures with lower false negatives. The plain model performed better than the 2-levels model, although the first level, i.e. a binary classification, between nevi and non-nevi yielded the best outcomes. Keywords Image processing · Deep learning · Classification · Skin cancer · Melanoma

1 Introduction Skin alterations are caused due to multiple factors, like allergies, infections, exposition to the sun, etc. The last one is a common practice of most people, who looks for a tan of their skin. However, this search for beauty can have a negative effect on the appearance of skin lesions. This is a typical example of one of the reasons for skin cancer.

B

Karl Thurnhofer-Hemsi [email protected] Enrique Domínguez [email protected]

1

Department of Computer Languages and Computer Sciences, University of Málaga, Boulevar Louis Pasteur, 35, 29071 Málaga, Spain

2

Biomedical Research Institute of Málaga (IBIMA), C/ Doctor Miguel Díaz Recio, 28, 29010 Málaga, Spain

123

K. Thurnhofer-Hemsi, E. Domínguez

Melanoma and non-melanoma skin cancer are highly present in Caucasians. The most common non-melanoma affections are basal cell carcinoma and squamous cell carcinoma. There were more than one million cases in 2018, being the 5th most common cancer. On the other side, melanoma is a less occurring cancer (in the 19th position), with around three hundred thousand new cases last year. Despite the lower number of detections, melanoma causes most of the mortality cases within the skin cancer area. Melanoma is caused by an abnormal multiplication of melanocytes, the cells that produce pigment and give color to the skin. The sooner the melanoma is detected, the greater the chances of cure. Nevertheless, it could spread to other parts of the body if it is not detected early [1], causing an irremediable effect. The problem resides in the capacity of the detection of melanomas, as they are similar in characteri