Deep Learning Ensemble Methods for Skin Lesion Analysis towards Melanoma Detection

Ali, Redha; Hardie, Russell C.; Narayanan, Barath Narayanan; Silva, Supun De

doi:10.1109/naecon46414.2019.9058245

Cited by 38 publications

(27 citation statements)

References 21 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…In this section, we compare the segmentation result of our models with some recent network models in terms of DSC, F1, and mIoU metrics. In addition to Unet and Unet++, the models used to comparison in this study include the Unet-based models (Double-Unet [ 39 ], R2U-Net [ 32 ], CU-Net [ 38 ], Multi-ResUnet [ 36 ], Cascade U-Resnet [ 10 ]), and others architecture networks (Deeplab V3+ [ 55 ], Generative Adversarial Network (GAN) [ 56 ]). Table 5 compares the proposed model results with the current models.…”

Section: Resultsmentioning

confidence: 99%

TMD-Unet: Triple-Unet with Multi-Scale Input Features and Dense Skip Connection for Medical Image Segmentation

et al. 2021

View full text Add to dashboard Cite

Deep learning is one of the most effective approaches to medical image processing applications. Network models are being studied more and more for medical image segmentation challenges. The encoder–decoder structure is achieving great success, in particular the Unet architecture, which is used as a baseline architecture for the medical image segmentation networks. Traditional Unet and Unet-based networks still have a limitation that is not able to fully exploit the output features of the convolutional units in the node. In this study, we proposed a new network model named TMD-Unet, which had three main enhancements in comparison with Unet: (1) modifying the interconnection of the network node, (2) using dilated convolution instead of the standard convolution, and (3) integrating the multi-scale input features on the input side of the model and applying a dense skip connection instead of a regular skip connection. Our experiments were performed on seven datasets, including many different medical image modalities such as colonoscopy, electron microscopy (EM), dermoscopy, computed tomography (CT), and magnetic resonance imaging (MRI). The segmentation applications implemented in the paper include EM, nuclei, polyp, skin lesion, left atrium, spleen, and liver segmentation. The dice score of our proposed models achieved 96.43% for liver segmentation, 95.51% for spleen segmentation, 92.65% for polyp segmentation, 94.11% for EM segmentation, 92.49% for nuclei segmentation, 91.81% for left atrium segmentation, and 87.27% for skin lesion segmentation. The experimental results showed that the proposed model was superior to the popular models for all seven applications, which demonstrates the high generality of the proposed model.

show abstract

Section: Resultsmentioning

confidence: 99%

TMD-Unet: Triple-Unet with Multi-Scale Input Features and Dense Skip Connection for Medical Image Segmentation

et al. 2021

View full text Add to dashboard Cite

show abstract

“…[31] Image enhancement using mean subtraction and standard deviation-based normalization. [32] Artifact removal and image enhancement using color constancy with shades of gray. [33] Artifact removal and image enhancement using histogram-based preprocessing.…”

Section: Studymentioning

confidence: 99%

“…Myriads of image segmentation methods have been proposed in the literature for skin lesions. They include thresholding [ 17 , 19 ], clustering [ 20 , 21 , 22 , 23 ], statistical region merging [ 8 , 10 ], saliency [ 19 , 24 , 25 , 26 , 27 , 28 , 29 ], and deep learning [ 2 , 30 , 31 , 32 , 33 , 34 , 35 ]. Although multifarious image segmentation methods exist in the literature, accurate segmentation of skin lesions is still a challenging open problem because of the heterogeneous properties of dermoscopic images.…”

Section: Introductionmentioning

confidence: 99%

“…The asymmetric features such as the irregular border, uneven shape, and colors of skin lesions can cause difficulties in accurately segregating the regions of skin lesions from the background. Moreover, images with low contrast between skin lesions and normal skin, the presence of undesirable artifacts such as color charts, marker ink, ruler marks, air bubbles, vignette, noise, and inherent cutaneous artifacts such as hair and blood vessels can make the segmentation obligation more challenging [ 3 , 5 , 6 , 12 , 16 , 32 , 36 , 37 ]. The artificial correlation created by these undesirable artifacts could adversely affect the performance of segmentation methods for skin lesions.…”

Section: Introductionmentioning

confidence: 99%

See 1 more Smart Citation

Preprocessing Effects on Performance of Skin Lesion Saliency Segmentation

Joseph

Olugbara

2022

Diagnostics

View full text Add to dashboard Cite

Despite the recent advances in immune therapies, melanoma remains one of the deadliest and most difficult skin cancers to treat. Literature reports that multifarious driver oncogenes with tumor suppressor genes are responsible for melanoma progression and its complexity can be demonstrated by alterations in expression with signaling cascades. However, a further improvement in the therapeutic outcomes of the disease is highly anticipated with the aid of humanoid assistive technologies that are nowadays touted as a superlative alternative for the clinical diagnosis of diseases. The development of the projected technology-assistive diagnostics will be based on the innovations of medical imaging, artificial intelligence, and humanoid robots. Segmentation of skin lesions in dermoscopic images is an important requisite component of such a breakthrough innovation for an accurate melanoma diagnosis. However, most of the existing segmentation methods tend to perform poorly on dermoscopic images with undesirable heterogeneous properties. Novel image segmentation methods are aimed to address these undesirable heterogeneous properties of skin lesions with the help of image preprocessing methods. Nevertheless, these methods come with the extra cost of computational complexity and their performances are highly dependent on the preprocessing methods used to alleviate the deteriorating effects of the inherent artifacts. The overarching objective of this study is to investigate the effects of image preprocessing on the performance of a saliency segmentation method for skin lesions. The resulting method from the collaboration of color histogram clustering with Otsu thresholding is applied to demonstrate that preprocessing can be abolished in the saliency segmentation of skin lesions in dermoscopic images with heterogeneous properties. The color histogram clustering is used to automatically determine the initial clusters that represent homogenous regions in an input image. Subsequently, a saliency map is computed by agglutinating color contrast, contrast ratio, spatial feature, and central prior to efficiently detect regions of skin lesions in dermoscopic images. The final stage of the segmentation process is accomplished by applying Otsu thresholding followed by morphological analysis to obliterate the undesirable artifacts that may be present at the saliency detection stage. Extensive experiments were conducted on the available benchmarking datasets to validate the performance of the segmentation method. Experimental results generally indicate that it is passable to segment skin lesions in dermoscopic images without preprocessing because the applied segmentation method is ferociously competitive with each of the numerous leading supervised and unsupervised segmentation methods investigated in this study.

show abstract

“…Skin lesion diagnosis by using deep learning methods is one of these popular field of interests (Hosny et al, 2019;Mahbod et al, 2019, may;Ray et al, 2020, may;Zhang et al, 2019). In particular, melanoma classification with deep learning methods gained popularity in last years (Ali et al, 2019, july;Favole et al, 2020, may;Winkler et al, 2019;.…”

Section: Deep Learning Models For Benign and Malignant Skin Lesion CLmentioning

confidence: 99%

A modified version of GoogLeNet for melanoma diagnosis

Yılmaz

Trocan

2021

Journal of Information and Telecommunication

View full text Add to dashboard Cite

Differential diagnosis of malignant melanoma, which is the cause of more than 75% of deaths amongst skin lesions, is vital for patients. Artificial intelligence-based decision support systems developed for the analysis of medical images are in the solution of such problems. In recent years, various deep learning algorithms have been developed to be used for this purpose. In our previous study, we compared the performances of AlexNet, GoogLeNet and ResNet-50 for the differential diagnosis of benign and malignant melanoma on International Skin Imaging Collaboration: Melanoma Project (ISIC) dataset. In this study, we proposed a CNN model by modifying the GoogLeNet algorithm and we compared the performance of this model with the previous results. For the experiments, we used 19,373 benign and 2197 malignant diagnosed dermoscopy images obtained from this public archive. We compared the performance results according to the eight different performance metrics including polygon area metric (PAM), classification accuracy (CA), sensitivity (SE), specificity (SP), area under curve (AUC), kappa (K), F measure metric (FM) and time complexity (TC) measures. According to the results, our proposed CNN achieved the best classification accuracy with 0.9309 and decreased the time complexity of GoogLeNet from 283 min 50 to 256 min 26 s.

show abstract

Deep Learning Ensemble Methods for Skin Lesion Analysis towards Melanoma Detection

Cited by 38 publications

References 21 publications

TMD-Unet: Triple-Unet with Multi-Scale Input Features and Dense Skip Connection for Medical Image Segmentation

TMD-Unet: Triple-Unet with Multi-Scale Input Features and Dense Skip Connection for Medical Image Segmentation

Preprocessing Effects on Performance of Skin Lesion Saliency Segmentation

A modified version of GoogLeNet for melanoma diagnosis

Contact Info

Product

Resources

About