Research Article
Contributions and Limitations About the Use of Deep Learning for Skin Diagnosis: A Review
@INPROCEEDINGS{10.1007/978-3-031-22324-2_11, author={Eduardo L. L. Nascimento and Angel Freddy Godoy Viera}, title={Contributions and Limitations About the Use of Deep Learning for Skin Diagnosis: A Review}, proceedings={Data and Information in Online Environments. Third EAI International Conference, DIONE 2022, Virtual Event, July 28-29, 2022, Proceedings}, proceedings_a={DIONE}, year={2022}, month={12}, keywords={Deep learning Skin lesion classification Skin lesion diagnostics Skin disease Dermatopathology Image recognition}, doi={10.1007/978-3-031-22324-2_11} }- Eduardo L. L. Nascimento
Angel Freddy Godoy Viera
Year: 2022
Contributions and Limitations About the Use of Deep Learning for Skin Diagnosis: A Review
DIONE
Springer
DOI: 10.1007/978-3-031-22324-2_11
Abstract
The aim of this study is to analyze the characteristics and applicability of Deep Learning (DL) models for the diagnosis of skin diseases. This study is characterized as a bibliographic review, exploratory-descriptive, qualitative in nature. Primary data was reported in the article databases. A total of 37 articles were analyzed to characterize the use of DL for the diagnosis of skin diseases. The survey results that public datasets access is mostly used in these surveys are (86%). The data collection that stood out was ISIC - International Skin Imaging Collaboration (54%). Greater commonly used data types in these models are images. Ultimately used model is the Convolutional Neural Network (CNN) and the uttermost used pre-trained model was ResNet. The most used techniques in the articles, in addition to classification (73%), focused on data segmentation (35%) and feature extraction (24%). The evaluation indicators that stand out are accuracy (89%), sensitivity (75%), and specificity (67%). The literature indicated that the approaches of studies that use DL for classification of skin diseases are very promising, however, practically all of the applied technologies have a greater need for interaction with clinical practices. As a suggestion for different works, studies that approach the task of DL work for diagnosis of different ethnic groups and their solutions for the democratization of such technologies.
