Early Prediction of Melanoma Using Image Processing and Conventional Neural Network

Abstract

Skin cancer is quite a common type of cancer. Its incidence is higher in Caucasians, and melanoma is the most lethal type. To increase patient prognosis, developing tools to assist in early diagnosis is necessary. In this work, we develop an easy and accessible mobile app to assist melanoma detection. The app is linked to a classification model for classifying the images as melanoma or non-melanoma, motivated by the performance of convolutional neural networks in computer vision trained on images collected from smartphones and clinical lesion information. The significance of a melanoma detection application is in its capacity to facilitate early detection, which is done by collecting a comprehensive and representative dataset of skin lesion photos. These photos encompass a wide range of diversity and balance, encompassing both melanoma and non-melanoma instances. Moreover, it provides annotations on the photos by adding labels that indicate whether each lesion is classified as melanoma or non-melanoma. Since the occurrence of melanoma is much smaller than other skin lesions, most of the datasets for this problem are imbalanced which has been demonstrated by the findings of this study. However, achieving data set balance is a crucial step in ensuring optimal performance of the detection model for both melanoma and non-melanoma instances. Nevertheless, it is of utmost importance to uphold an accurate portrayal of the frequency of melanoma in practical situations and to further enhance the model's efficacy by engaging in partnerships with dermatologists and persistently gathering data. In order to come up with a balanced dataset, an approach has been presented that is based on evolutionary algorithm and it entails enhancing the data distribution. The findings suggested that evolutionary algorithms have the potential to aid in the selection of the most pertinent features from photos, hence enhancing the accuracy of the detection model.