Melanoma Skin Cancer Detection Application using Convolutional Neural Network (CNN)

Abstract

Introduction: The skin, the body's largest organ, protects against external threats, notably harmful ultraviolet (UV) radiation, excessive exposure to which can cause melanoma, a potentially fatal skin cancer.

Objectives: This study developed a melanoma detection model by training a machine learning algorithm on a Kaggle image dataset, assessing its accuracy, and then integrating it into a mobile app for user-friendly skin image analysis.

Methods: This study developed a deep Convolutional Neural Network (CNN) to accurately detect melanoma in dermoscopic images with 5,000 classified as benign and 4,605 as malignant. The model's performance was evaluated on an independent test set of 1,000 images, equally distributed between benign and malignant classes.  

Results: The result shows that training and validation accuracy improved over 20 epochs (starting at 0.6472 and 0.8409, respectively, and reaching over 0.90, converging to 0.9044 and 0.9096), while training and validation loss decreased (from 0.6548 and 0.4521 to 0.2235 and 0.2806), respectively.  

Conclusions: A trained machine learning model was optimized for mobile deployment by converting it to TensorFlow Lite and then integrated into an Android application developed with Android Studio. This study demonstrates a reliable model for melanoma prediction, successfully implemented in a mobile application for improved early diagnosis. Future work should focus on enhancing the model's performance through expanded datasets and alternative algorithms.