Deep features to classify skin lesions

Kawahara, Jeremy; BenTaieb, Aïcha; Hamarneh, Ghassan

doi:10.1109/isbi.2016.7493528

Cited by 253 publications

(180 citation statements)

References 8 publications

Supporting

Mentioning

158

Contrasting

Unclassified

Order By: Relevance

“…We also run a single comparison with the VGG-16 model [18] to evaluate the impact of that deeper (and more expensive) architecture. Because the original implementations are in MATLAB, which was not convenient for our computational infrastructure, we reimplemented those models in Lasagne 5 and Nolearn 6 . In the experiments with transfer learning, we get the source networks pre-trained on ImageNet, train them from scratch on Retinopathy, or fine-tune on Retinopathy the model pre-trained on ImageNet.…”

Section: Methodsmentioning

confidence: 99%

“…Works employing DNNs for melanoma screening either train a network from scratch [6,7,9], or transfer knowledge from ImageNet [4,5,8,10,11] the choice of DNN architecture and implementation frameworkthe most common framework is Caffe [4,8,11], and the most common architectures are ResNet [10], DRN-101 [11], AlexNet [5], and VGG-16 [8]. Schemes for artificially augmenting training data, and for transferring learned knowledge also vary.…”

Section: Introductionmentioning

confidence: 99%

“…Schemes for artificially augmenting training data, and for transferring learned knowledge also vary. Segmenting the lesion before feature extraction is a common choice for melanoma screening [8,11], although DNNs generally allow forgoing it [1,4,5,8,10,14]. In this paper we forgo all preprocessing steps, like lesion segmentation, or artifacts/hair removal.…”

Section: Introductionmentioning

confidence: 99%

See 2 more Smart Citations

Knowledge transfer for melanoma screening with deep learning

Menegola

Fornaciali

Pires

et al. 2017

2017 IEEE 14th International Symposium on Biomedical Imaging (ISBI 2017)

163

View full text Add to dashboard Cite

Knowledge transfer impacts the performance of deep learning -the state of the art for image classification tasks, including automated melanoma screening. Deep learning's greed for large amounts of training data poses a challenge for medical tasks, which we can alleviate by recycling knowledge from models trained on different tasks, in a scheme called transfer learning. Although much of the best art on automated melanoma screening employs some form of transfer learning, a systematic evaluation was missing. Here we investigate the presence of transfer, from which task the transfer is sourced, and the application of fine tuning (i.e., retraining of the deep learning model after transfer). We also test the impact of picking deeper (and more expensive) models. Our results favor deeper models, pretrained over ImageNet, with fine-tuning, reaching an AUC of 80.7% and 84.5% for the two skin-lesion datasets evaluated.

show abstract

Section: Methodsmentioning

confidence: 99%

Section: Introductionmentioning

confidence: 99%

Section: Introductionmentioning

confidence: 99%

See 1 more Smart Citation

Knowledge transfer for melanoma screening with deep learning

Menegola

Fornaciali

Pires

et al. 2017

2017 IEEE 14th International Symposium on Biomedical Imaging (ISBI 2017)

163

View full text Add to dashboard Cite

show abstract

“…Codella et al said that to integrate CNNs(convolutional neural network), sparse coding and support vector machine (SVM [40], [41], [42], [43], [44]) for melanoma recognition [18]. Next Kawahara et al presented a fully convolutional neural network based on AlexNet 19] to extract representative features of melanoma [20]. But these methods either just depend on the features trained from natural image dataset (such as ImageNet [21]) without sufficiently considering the characteristics of melanoma or utilize CNNs with quite shallow architecture.…”

Section: Literature Surveymentioning

confidence: 99%

Design and Implementation of CNN for detecting Melanoma through Image Processing

Ranjan¹

2018

IJRASET

View full text Add to dashboard Cite

Melanoma is a type of skin cancer which is caused by prolonged exposure to u-v light. Early days melanoma detected through dermoscopy where doctor tries to recognize melanoma based on his experience and assumption. But many cases patient who suffers from melanoma dies because of it not detected early stages. In recent years, several procedures have been proposed to detect and analyze skin cancer. Early stages only skin cancer can be classified as malignant or benign. All proposed methods contribute better values of precision, sensitivity, specificity, and accuracy.

show abstract

“…By using of deep convolutional neural networks (DC-NN) machine learning algorithm in [12], the authors developed a three pattern detectors approach on a set of 211 images and reported accuracy below than 85%. The CNN model used in [13] to extract features with pooling techniques to recognize PSLS skin lesions and achieved 85.8% accuracy. The deep-neuralnetwork (DNN) is used to classify melanoma and achieved 89.3% accuracy.…”

Section: Introductionmentioning

confidence: 99%