Abstract
Skin disease is perhaps the most widely recognized malignant growth, and it creates because of an assortment of dermatological conditions. It is divided into numerous categories depending on morphological characteristics, color, structure, and texture. For patients with skin cancer to have a low death rate, early and prompt detection and diagnosis of malignant skin cancer cells is crucial. Dermoscopic pictures currently have image quality restrictions, such as color, artefact, and blur, which can make diagnosis difficult. Early and accurate diagnosis of skin disorders is challenging due to cost, effort, overfitting, larger feature dimension, and low detection accuracy. To address these concerns, we intend to present the most effective Network Model for classifying skin diseases. The skin lesion zones are then efficiently segmented using a new segmentation procedure. This review/survey paper will discuss the recent research on skin lesion classification using Deep Learning.
Access provided by Autonomous University of Puebla. Download conference paper PDF
Similar content being viewed by others
Keywords
- Skin lesion classification
- Segmentation
- Convolutional network model
- Deep learning
- Convolutional neural network
1 Introduction
When some cells in the body grow too quickly, they can spread to other places in the body. This is called “cancer.” Each human body is made up of up to a million cells. A cancerous condition can develop anywhere any of these cells are located. Human cells create and isolate (called cell division) to make new cells as the body needs them. Malignant growth happens when this typical framework is screwed up by hereditary issues. Cells begin to develop at a wild rate. Some of these cells can work together to make a tumour, which is a type of growth. It could be cancerous or not cancerous. One that is malignant is one that can grow and spread all over the body. In medicine, a benign tumour is one that has the chance to grow but not spread. A condition called skin cancer is when abnormal cells in the epidermis, the outer layer of the skin, grow too quickly. This is caused by damage to DNA that hasn’t been repaired. These changes make skin cells grow quickly, which leads to tumour. The four major kinds of skin cancer are BCC, SCC, melanoma, and Merkel cell carcinoma (MCC); however, there also exist many kinds of skin cancers.
2 Skin Cancer
As per the W.H.O, there were 1,329,779 new instances of skin malignant growth in 2018. Melanoma kills many individuals who pass on from skin disease. This year, about 287,700 people are predicted to get melanoma. Nearly 6700 deaths are expected. The number of skin cancer deaths doesn’t matter. Even with high-level treatment, advanced stage malignant melanoma has a 15% survival probability if found early.
2.1 Deep Learning (DL) Techniques
DL has exhibited promising result in segmenting skin infections, a tough topic in computer vision. Various grounded models of deep literacy have been presented. These techniques give excellent skin lesion segmentation results. These architectures include DCNNs, U-Nets, Fully Convolutional Neural Networks (FCNs), Deep Fully Convolutional Neural Networks (FCNNs), and Convolutional DE-Convolutional Neural Networks (CDCNN). Table 1 shows the differences between networks and their use in deep learning.
3 Dataset
A dataset (likewise spelled ‘informational index’) is an assortment of crude measurements and data created by an exploration study. The motivation behind Datasets is to extract from straightforwardly speaking with the data set utilizing basic SQL explanations. The Dataset’s objective is to serve as a small local replica of the data that you bothered about so that you don’t have to make costly high-inertia database decisions in the future. A Dataset is the most fundamental information container in PyMVPA. It is the most basic sort of data storage, as well as a standard compartment storing the outputs of most algorithms. As an illustration. An informational index is a collection of numbers or characteristics that pertain to a specific topic. A dataset is made up of the grades of all students in a certain class, for example. The number of fish ingested by each dolphin in an aquarium is referred to as a dataset. By pointing to the article, the publicly available informational index used for skin characterization, division, and discovery is shown in Table 2. The majority of the analysts used a range of datasets to develop their own models and calculations.
4 Distribution of Classes
The below Fig. 1 easily states that the distribution of classes according to the dataset and experimenter frequently and infrequently used the dataset. Most of the experimenter used the combination of one or further dataset to classify and member the skin lesion and used the dataset with some other armature or transfer literacy frame to achieve their delicacy.
5 Literature Survey
Predicted and investigated earlier work from 2021-2020 in the Table 3. The table below summarises our findings. The table comprises four columns: the reference title, first author, and publication year, the purpose, the method or model employed, and the dataset, and the significant results uncovered when analysing the papers submitted. Several methodologies and algorithms were used to obtain its accuracy. At the end of each investigation, we realised that the network model requires very few datasets. This was done via pixelwise correlation in [2] to sharpen the irregularity and segment the skin lesion. Rather than changing organization boundaries, they utilize Dual Encoder Architecture to work collaboratively. Their dilatation rate differs from ADAM’s. The solution outperforms U-Net in both the ISBI2017 and ISIC2018 datasets, highlighting the usefulness of both the dual encoder architecture as well as the adaptive dual attention module in enhancing the segmentation process of skin lesion. [3] In order to screen melanoma, the lesions changes are monitored in short time frame. The global features are extracted from dermoscopy by proposing tensorial neural network. They used a spatial transformer network to improve detection performance and invented the SegLoss regularisation factor. [1] Segmenting and categorising skin lesions. In this, the MB-DCNN method performs both the operations such as segmentation as well as classification. This method involves coarse segmentation network, an updated segmentation network (enhanced-SN) and a mask-guided classification network. This processing creates class imbalance as well as hard easy pixel imbalance. [4] In this study, uneven learning focuses on smaller or more complicated minority courses. The proposed SPBL (Self-Paced Balance Learning) method tries to generate a balanced representation of all classes. [5] According to the authors, a multistage unit-vise deep dense residual network along with transition and supervision blocks can support dense and minimal skip residual learning. Each layer may analyse the previous layers’ features locally and less difficultly than its counter network.
6 Skin Lesion Detection in Melanoma Skin Cancer
The ABCD technique can detect melanoma skin cancer, which is lethal. The ABCD rule helps physicians, nurses, and patients identify melanoma-like skin lesions. Melanoma can appear suddenly on the skin. In or near a mole or other dark area on the skin. That’s why knowing your moles’ colour, size, and placement on your body is critical to spotting changes (Fig. 2).
Asymmetry means that one half of the mole does not match the other in the ABCD melanoma detection criteria. The mole’s border and margins are not smooth. The mole is multicolored, hence the C. The letter D denotes a larger diameter or width than a pencil eraser. Compared to a normal mole, E stands for evolving size, shape, and colour. The Table 4 summarizes the many methods for detecting Melanoma skin lesions.
7 Skin Lesion Segmentation in Melanoma Skin Cancer
Numerous approaches are already developed and applied for the efficient segmentation of skin lesions. Particularly in recent times, the deep learning approach, Convolutional Neural Network (CNNs) has achieved veritably successful results in segmentation of skin lesions Still, CNNs accepts low- resolution images for dwindling the number of computations and parameters in the network. This situation may lead to the loss of some important features in the image. Table 5 describes the methodology and approaches used to segment melanoma skin lesion.
8 Classification in Melanoma
From once exploration in this field, it’s apparent that CNN has an extraordinary capability to perform skin lesion bracket in competition with professional dermatologists (Table 6). In fact, there have been cases where CNN has outperformed professional dermatologists as well in Fig. 3. Skin lesions can be classified in two ways, according to CNN. A CNN is utilized to extract picture points in the first case, while another classifier classifies the images within seconds. In other circumstance, CNN is used to conduct end-to-end literacy. Scrape literacy and pretrained model literacy are two types of scrape literacy. To overcome the problem of overfitting, more photographs are essential to train the CNN from scraping. Training CNN from scrape is more difficult due to the small number of skin lesion images necessary for training. A better method is to learn from a pre-trained system which is Transfer Literacy (TL). The concept attribute is introduced to the trained model by TL, which helps the model learn better with little input.
9 Conclusion
Deep learning’s capacity to handle vast amounts of features makes it useful for unstructured data. We analyzed more current papers and found that most researchers employed neural network techniques with deep learning. For example: Convolutional Neutral Networks and Recurrent Neural Networks. In a CNN, the hidden layers don’t necessarily share their output with the next layer (known as convolutional layers). Automated feature extraction from images using deep learning. They can learn what to search for in images by analyzing many images. As shown above, staggered Profound Learning models are particularly useful in distinguishing confounded data from input images. Convolutional neural networks can also drastically reduce computation time by utilizing GPUs, which many do not use. In this way, the grouping model can recognize the image. Keras has inbuilt capacities that make it simple to customize and make a neural organization with CNN design. We plan to create our own Convolutional Neutral Network for Melanoma Order with our own rendition of turns.
References
Wu, H., Pan, J., Li, Z., Wen, Z., Qin, J.: Automated skin lesion segmentation via an adaptive dual attention module. IEEE Trans. Med. Imaging 40(1), 357–370 (2020). https://doi.org/10.1109/TMI.2020.3027341
Zhang, B., et al.: Short-term lesion change detection for melanoma screening with novel siamese neural network. IEEE Trans. Med. Imaging 40(3), 840–851 (2020). https://doi.org/10.1109/TMI.2020.3037761
Xie, Y., Zhang, J., Xia, Y., Shen, C.: A mutual bootstrapping model for automated skin lesion segmentation and classification. IEEE Trans. Med. Imag. 39(7), 2482–2493 (2020). https://doi.org/10.1109/TMI.2020.2972964
Yang, J., et al.: Self-paced balance learning for clinical skin disease recognition. IEEE Trans. Neural Netw. Learning Syst. 31(8), 2832–2846 (2020). https://doi.org/10.1109/TNNLS.2019.2917524
Gessert, N., et al.: Skin lesion classification using CNNS with patch-based attention and diagnosis-guided loss weighting. IEEE Trans. Biomed. Eng. 67(2), 495–503 (2019)
Razzak, I., Naz, S.: Unit-vise: deep shallow unit-vise residual neural networks with transition layer for expert level skin cancer classification. IEEE/ACM Trans. Comput. Biol. Bioinform. 19(02), 225–1234 (2020)
Wang, S., Yin, Y., Wang, D., Wang, Y., Jin, Y.: Interpretability-based multimodal convolutional neural networks for skin lesion diagnosis. IEEE Trans. Cybern. 52(12), 12623–12637 (2022)
Zhang, J., Xie, Y., Xia, Y., Shen, Y.: Attention residual learning for skin lesion classification. IEEE Trans. Med. Imaging 38(9), 2092–2103 (2019)
Wang, X., Jiang, X., Ding, H., Liu, J.: Bi-directional dermoscopic feature learning and multi-scale consistent decision fusion for skin lesion segmentation. IEEE Trans. Image Process. 29, 3039–3051 (2019)
Yu, Z., et al.: Melanoma recognition in dermoscopy images via aggregated deep convolutional features. IEEE Trans. Biomed. Eng. 66(4), 1006–1016 (2018)
Back, S., et al.: Robust skin disease classification by distilling deep neural network ensemble for the mobile diagnosis of herpes zoster. IEEE Access 9, 20156–20169 (2021). https://doi.org/10.1109/ACCESS.2021.3054403
Jiang, S., Li, H., Jin, Z.: A visually interpretable deep learning framework for histopathological Image-based skin cancer diagnosis. IEEE J. Biomed. Health Inform. 25(5), 1483–1494 (2021)
Gong, A., Yao, X., Lin, W.: Dermoscopy image classification based on StyleGAN and DenseNet201. IEEE Access 9, 8659–8679 (2021)
Thurnhofer-Hemsi, K., López-Rubio, E., Domínguez, E., Elizondo, D.A.: Skin lesion classification by ensembles of deep convolutional networks and regularly spaced shifting. IEEE Access 9, 112193–112205 (2021)
Bian, J., Zhang, S., Wang, S., Zhang, J., Guo, J.: Skin lesion classification by multi-view filtered transfer learning. IEEE Access 9, 66052–66061 (2021)
Adegun, A.A., Viriri, S.: Deep learning-based system for automatic melanoma detection. IEEE Access 8, 7160–7172 (2019)
Ahmad, B., Usama, M., Huang, C.M., Hwang, K., Hossain, M.S., Muhammad, G.: Discriminative feature learning for skin disease classification using deep convolutional neural network. IEEE Access 8, 39025–39033 (2020)
Pham, T.C., Doucet, A., Luong, C.M., Tran, C.T., Hoang, V.D.: Improving skin-disease classification based on customized loss function combined with balanced mini-batch logic and real-time image augmentation. IEEE Access 8, 150725–150737 (2020)
Fan, X., et al.: Effect of image noise on the classification of skin lesions using deep convolutional neural networks. Tsinghua Sci. Technol. 25(3), 425–434 (2019)
Goyal, M., Oakley, A., Bansa, P., Dancey, D., Yap, M.H.: Skin lesion segmentation in dermoscopic images with ensemble deep learning methods. IEEE Access 8, 4171–4181 (2019)
Zhang, G., et al.: DSM: A deep supervised multi-scale network learning for skin cancer segmentation. IEEE Access 7, 140936–140945 (2019)
Song, L., Lin, J., Wang, Z.J., Wang, H.: An end-to-end multi-task deep learning framework for skin lesion analysis. IEEE J. Biomed. Health Inform. 24(10), 2912–2921 (2020)
Adegun, A.A., Viriri, S.: FCN-based DenseNet framework for automated detection and classification of skin lesions in dermoscopy images. IEEE Access 8, 150377–150396 (2020)
Tang, P., Liang, Q., Yan, X., Xiang, S., Zhang, D.: Gp-cnn-dtel: global-part cnn model with data-transformed ensemble learning for skin lesion classification. IEEE J. Biomed. Health Inform. 24(10), 2870–2882 (2020). https://doi.org/10.1109/JBHI.2020.2977013
Gu, Y., Ge, Z., Bonnington, C.P., Zhou, J.: Progressive transfer learning and adversarial domain adaptation for cross-domain skin disease classification. IEEE J. Biomed. Health Inform. 24(5), 1379–1393 (2019)
Author information
Authors and Affiliations
Corresponding authors
Editor information
Editors and Affiliations
Rights and permissions
Copyright information
© 2023 The Author(s), under exclusive license to Springer Nature Switzerland AG
About this paper
Cite this paper
Vidhyalakshmi, A.M., Kanchana, M. (2023). A Review on Detection and Diagnosis of Melanoma Carcinoma Using Deep Learning. In: Kottursamy, K., Bashir, A.K., Kose, U., Uthra, A. (eds) Deep Sciences for Computing and Communications. IconDeepCom 2022. Communications in Computer and Information Science, vol 1719. Springer, Cham. https://doi.org/10.1007/978-3-031-27622-4_19
Download citation
DOI: https://doi.org/10.1007/978-3-031-27622-4_19
Published:
Publisher Name: Springer, Cham
Print ISBN: 978-3-031-27621-7
Online ISBN: 978-3-031-27622-4
eBook Packages: Computer ScienceComputer Science (R0)