Melanoma Skin Cancer Detection based on Image Processing

Abstract

Background: Skin cancer is one of the most common forms of cancers among humans. It can be classified as non-melanoma and melanoma. Although melanomas are less common than non-melanomas, the former is the most common cause of mortality. Therefore, it becomes necessary to develop a Computer-aided Diagnosis (CAD) aiming to detect this kind of lesion and enable the diagnosis of the disease at an early stage in order to augment the patient’s survival likelihood. Aims: This paper aims to develop a simple method capable of detecting and classifying skin lesions using dermoscopy images based on ABCD rules. Methods: The proposed approach follows four steps. 1) The preprocessing stage consists of filtering and contrast enhancing algorithms. 2) The segmentation stage aims at detecting the lesion. 3) The feature extraction stage based on the calculation of the four parameters which are asymmetry, border irregularity, color and diameter. 4) The classification stage based on the summation of the four extracted parameters multiplied by their weights yields the total dermoscopy value (TDV); hence, the lesion is classified into benign, suspicious or malignant. The proposed approach is implemented in the MATLAB environment and the experiment is based on PH2 database containing suspicious melanoma skin cancer. Results and Conclusion: Based on the experiment, the accuracy of the developed approach is 90%, which reflects its reliability.

Overdiagnosis of melanoma – causes, consequences and solutions

“…For successful integration of an AI‐based algorithm into clinical or pathological routine, there must at least be the possibility of classifying a lesion as neither melanoma nor nevus, but for example as an “unknown category” [61]. In this respect, efforts are currently underway [62], for example with the International Skin Imaging Collaboration (ISIC) competition for the classification of melanomas (https://challenge2019.isic-archive.com/leaderboard.html). Moreover, categories of additional, specific skin diagnoses are imaged in developed algorithms [58,63].…”

Cilt Kanseri Teşhisi için Konvolüsyonel Sinir Ağları Tabanlı Bilgisayar Destekli Tanıda (CNN-CAD) Dijital Görüntü Kalitesinin İyileştirilmesi

“…According to the research, there are several methods currently used for pre-processing in skin cancer diagnosis. Some of them are represented in [20][21]. In this study, we used one most popular ones, which can be defined through 3 steps: Black-Hat transformation is used for hair removal, denoising and contrast enhancement are provided by Median Filter, while the image re-sizing is utilized to improve the quality and make all images one size.…”

Überdiagnose von Melanomen – Ursachen, Konsequenzen und Lösungsansätze

“…Soll ein KI‐basierter Algorithmus in die klinische beziehungsweise pathologische Routine erfolgreich integriert werden, muss mindestens die Möglichkeit bestehen, eine Läsion als weder Melanom noch Nävus einzuordnen, beispielsweise als „unbekannte Klasse“ [61]. Daran wird derzeit gearbeitet [62], beispielsweise im Wettbewerb der International Skin Imaging Collaboration (ISIC) zur Klassifikation von Melanomen (https://challenge2019.isic-archive.com/leaderboard.html). Zusätzlich werden Klassen zu spezifischen weiteren Haut‐Diagnosen in weiterentwickelten Algorithmen abgebildet [58,63].…”