Desarrollo de una aplicación web para detectar cáncer de piel utilizando aprendizaje profundo

Abstract

The development of advanced technologies in healthcare is one of the most exciting and challenging areas of the 21st century. The application of deep learning for medical diagnosis represents a significant advance in the ability of automated systems to assist professionals in making critical decisions. This work addresses a highly complex problem and has the potential to save lives by providing an accessible and accurate tool for the early detection of skin cancer. The development and implementation of a web application designed to detect skin cancer, using deep learning techniques, is detailed. The hypothesis posed infers that the use of a convolutional neural network will allow at least 90% accuracy in diagnosis to be achieved. Convolutional neural networks, specifically ResNet, were implemented with a data set of 25,331 images in total: 20,265 for training and 5,066 for testing.The CRISP-DM methodology used in this study includes data preprocessing using TensorFlow and evaluation of model performance using metrics such as accuracy, sensitivity, and loss. The results obtained revealed an accuracy of 94% and a sensitivity of 87%, exceeding the threshold of 90% accuracy established in the hypothesis. The analysis of the results indicates that the application complies with the hypothesis raised and the results are comparable to previous studies in the field, such as those carried out by Esteva et al. [1] and Haenssle et al. [2], who achieved accuracies of 72.1% and 74.2%, respectively. In conclusion, this work represents an important step towards the integration of advanced technologies in medicine, underlining the capacity of deep learning to transform the detection and treatment of skin cancer. Future research is suggested to improve accuracy, as well as to explore its applicability in other types of cancer.