MMMTsl – Multi-Modal\Tasking for Skin Lesion Classification using DNN


Classification of skin lesions using deep neural networks (DNN) has proved to deliver comparable results with medical practitioners. This work will be part of Derm.AI project and will use retrospective data of NHS from CTH (consulta a tempo e horas) to build the AI models to be incorporated in RSE-SIGA (a recent Integrated Access Management System). Through the DermAI project, FhP has access to patient metadata that is very useful to enhance the prediction of the skin lesions. Moreover, from the medical images, useful features can be extracted which can be used as metadata or outputs for multitasking training.



Direct impact in task execution of current and future projects. Main contributions will be related to Computer Vision and Explainable AI. The contributions of this work could be adapted for other similar projects where the attributes can be extracted and train a multi-modal and multi-task DNN.


Author: Rafaela Carvalho

Type: MSc thesis

Partner: FEUP – Faculdade de Engenharia da Universidade do Porto

Year: 2021