Skin Lesion Synthesis with Generative Adversarial Networks

Bissoto, Alceu; Perez, Fábio; Valle, Eduardo; Avila, Sandra

doi:10.1007/978-3-030-01201-4_32

Cited by 80 publications

(65 citation statements)

References 15 publications

Supporting

Mentioning

Contrasting

Unclassified

Order By: Relevance

“…Since both are publicly available, most of the recent works in skin lesion analysis rely on these datasets. When exploring reproducible works on lesion segmentation [8,24], dermoscopic attribute segmentation [14], skin lesion classification [7,19,23], or skin lesion synthesis [6] those two datasets are almost certain to be included. Next we describe their individual characteristics, and discuss how they differ.…”

Section: Datasetsmentioning

confidence: 99%

(De) Constructing Bias on Skin Lesion Datasets

Bissoto

Fornaciali

Valle

et al. 2019

2019 IEEE/CVF Conference on Computer Vision and Pattern Recognition Workshops (CVPRW)

Self Cite

View full text Add to dashboard Cite

Melanoma is the deadliest form of skin cancer. Automated skin lesion analysis plays an important role for early detection. Nowadays, the ISIC Archive and the Atlas of Dermoscopy dataset are the most employed skin lesion sources to benchmark deep-learning based tools. However, all datasets contain biases, often unintentional, due to how they were acquired and annotated. Those biases distort the performance of machine-learning models, creating spurious correlations that the models can unfairly exploit, or, contrarily destroying cogent correlations that the models could learn. In this paper, we propose a set of experiments that reveal both types of biases, positive and negative, in existing skin lesion datasets. Our results show that models can correctly classify skin lesion images without clinically-meaningful information: disturbingly, the machine-learning model learned over images where no information about the lesion remains, presents an accuracy above the AI benchmark curated with dermatologists' performances. That strongly suggests spurious correlations guiding the models. We fed models with additional clinically meaningful information, which failed to improve the results even slightly, suggesting the destruction of cogent correlations. Our main findings raise awareness of the limitations of models trained and evaluated in small datasets such as the ones we evaluated, and may suggest future guidelines for models intended for real-world deployment.

show abstract

Section: Datasetsmentioning

confidence: 99%

(De) Constructing Bias on Skin Lesion Datasets

Bissoto

Fornaciali

Valle

et al. 2019

2019 IEEE/CVF Conference on Computer Vision and Pattern Recognition Workshops (CVPRW)

Self Cite

View full text Add to dashboard Cite

show abstract

“…• Dois artigos de conferência [Bissoto et al 2018c, Bissoto et al 2019a] e um relatório técnico [Bissoto et al 2019b…”

Section: Resultados Da Pesquisaunclassified

“…Destacamos que esta contribuição foi publicada [Bissoto et al 2018c], e aindaé, até hoje, o estado da arte para a síntese de imagens de lesões de pele.…”

Section: Síntese De Imagens De Lesões De Peleunclassified

Análise de Lesões de Pele usando Redes Generativas Adversariais

Bissoto¹,

Avila²

2020

Anais Estendidos Do Simpósio Brasileiro De Computação Aplicada À Saúde (SBCAS 2020)

Self Cite

View full text Add to dashboard Cite

O câncer de pele é de longe o tipo mais comum de câncer. O diagnóstico precoce é fundamental para o tratamento do paciente, melhorando significativamente as taxas de sobrevida. O aprendizado profundo tornou-se o estado da arte na análise de lesões de pele, e os dados se tornaram um fator decisivo para impulsionar as soluções. O objetivo principal desta dissertação de mestrado é abordar dois problemas que surgem ao ter conjuntos de dados limitados: a incapacidade dos modelos de generalizar e o risco de viés do conjunto de dados. Investigamos os dois problemas separadamente, propondo métodos inovadores que contribuíram para o avanço da análise de lesões de pele.

show abstract

“…We implemented this augmentation through handcrafted image processing techniques, which may not be appropriate for producing reliable images. More advanced approaches, such as Generative Adversarial Networks or other generative architectures [2], might lead to better results.…”

Section: Resultsmentioning

confidence: 99%

Data Augmentation for Skin Lesion Analysis

Perez

Vasconcelos

Avila

et al. 2018

Lecture Notes in Computer Science

Self Cite

132

View full text Add to dashboard Cite

Deep learning models show remarkable results in automated skin lesion analysis. However, these models demand considerable amounts of data, while the availability of annotated skin lesion images is often limited. Data augmentation can expand the training dataset by transforming input images. In this work, we investigate the impact of 13 data augmentation scenarios for melanoma classification trained on three CNNs (Inception-v4, ResNet, and DenseNet). Scenarios include traditional color and geometric transforms, and more unusual augmentations such as elastic transforms, random erasing and a novel augmentation that mixes different lesions. We also explore the use of data augmentation at test-time and the impact of data augmentation on various dataset sizes. Our results confirm the importance of data augmentation in both training and testing and show that it can lead to more performance gains than obtaining new images. The best scenario results in an AUC of 0.882 for melanoma classification without using external data, outperforming the top-ranked submission (0.874) for the ISIC Challenge 2017, which was trained with additional data.

show abstract

Skin Lesion Synthesis with Generative Adversarial Networks

Cited by 80 publications

References 15 publications

(De) Constructing Bias on Skin Lesion Datasets

(De) Constructing Bias on Skin Lesion Datasets

Análise de Lesões de Pele usando Redes Generativas Adversariais

Data Augmentation for Skin Lesion Analysis

Contact Info

Product

Resources

About