Skin Cancer DetectionSkin Cancer Detection

Abstract

The skin is the largest organ and serves as a barrier to protect the internal organs, as well as the bones and muscles underneath it. When the skin deteriorates, the whole body is affected. Skin cancer is surprisingly common since the skin is so easily damaged by UV rays and other kinds of environmental pollutants. It seems that the two most common forms of skin cancer are benign melanoma and malignant melanoma. Malignant moles are significantly more harmful than benign moles and create bleeding blisters on the skin. They get their name because they resemble the more deadly melanocyte cells. Over 700,000 skin lesions are found annually in the United States, reports the American Cancer Society. Apollo and other groups have shown that melanoma is more common in those above the age of 41. Current techniques allow for the possible early identification of skin cancer. Saving lives and preventing further spread of skin cancer depend on prompt diagnosis. People with lighter skin tones, regardless of age, are at a higher risk. It will be tough for even the most seasoned dermatologist to diagnose skin cancer and estimate how far it will develop. As a consequence, several applications and programs came into being. Every year, millions of people lose their lives to skin cancer, making it the most lethal form of the disease. In order to maintain a low mortality rate and a high survival rate, it is crucial to diagnose potentially dangerous cases of skin cancer early and treat them accordingly. Relevant publications mostly focus on the use of machine learning in algorithm design. However, the accuracy and precision provided by these algorithms are not always perfect. Image noise is reduced during preprocessing with the use of sharpening and smoothing filters. The optimal performance of the system was maintained, and new insights were gained, by using a deep learning convolution neural network (DL-CNN) for the multi-class classification of skin cancer. Indeed, all nine forms of skin cancer may be properly categorized using the study\'s findings.

Introduction

I. INTRODUCTION

Skin cancer has been the most frequent cancer in the globe since the 1970s. In recent decades, there has been an increase in the diagnosis of both non-melanoma and melanoma skin cancers. While the World Health Organization (WHO) estimates that 1 in 3 people may get skin cancer at some point in their life, the Skin Cancer Foundation puts that figure at 1 in 5. In the Western Hemisphere, skin cancer has been on the increase for millennia. This trend is not exclusive to the US, Canada, or Australia; it can be seen in many other countries as well. The rapid spread and widespread impact of skin infections makes them a major public health concern. Skin cancer is responsible for 1.79 percent of the global disease burden in terms of disability-adjusted life years [1], according to a study published in 2017. Medicare in the United States lost more than $8 billion that year [2] because of the high expense of treating skin cancer, which accounts for around 7% of all newly diagnosed cancer cases worldwide. In spite of the fact that persons of color have a 20-30 fold increased chance of developing melanoma compared to those with lighter skin tones, the mortality risk has been shown to be either greater or lower depending on the kind of melanoma.

The key to effective treatment of a skin condition is a correct diagnosis. Cerebroscopy using this method may aid in the early detection of melanoma when it is easier to cure. Accurately diagnosing melanoma needs the skills of a dermatologist with relevant experience. Because malignant and non-malignant skin lesions seem identical and there is no clear demarcation between the two, melanoma may be difficult to identify. If a reliable automated method could be created to diagnose skin cancers by studying skin lesions, it would be a huge benefit to pathologists. In a time when information is becoming difficult to get, this is more crucial than ever. This study demonstrated the shortcomings of three popular categorization methods: K-nearest Neighbors, Support Vector Machines, and Decision Trees. After examining the mathematical underpinnings of classification, specialists concluded that deep learning models provided the most cutting-edge method of producing the desired outcomes. We tested out a broad array of mathematical models, some of which made use of learning algorithms and others of which did not. However, the quality and depth of activation were found to be insufficient in the pre-trained models. Together, we were able to combine our mathematical expertise and create a model called a Dense Convolutional Network, which has an accuracy of over 86.6%.

II. LITERATURE SURVEY

With the use of an ECOC support vector machine (SVM) and a deep convolutional neural network, Hameed et al. [3] developed a workaround. Using the proposed technique, pictures of skin lesions have been created, implemented, and assessed. It is now possible to label these images as either healthy, acne-prone, eczema-prone, benign, or malignant. There were 9,140 images from various sources included in the study. Alex NET, a pre-trained convolutional neural network model, was used to extract the features. For this specific classification task, we choose the ECOC SVM classifier. Accuracy improved to 86.21 percent with the help of ECOC SVM. A 10-fold cross-validation procedure was used to avoid overfitting the data. The results demonstrate that the retrieved characteristics from the convolutional neural network may help enhance the classification accuracy of a wide variety of skin lesions.

Aldhyani et al. [4] propose a CNN-based method that effectively employs kernels and activation functions. The proposed model used just 172,363 parameters but nevertheless managed an astonishing 97.85% accuracy on the test dataset. Data was segmented into seven classes for training the model. Disease classification may be aided by applying the proposed model on a dataset with additional categories. Benign keratosis, melanoma, and melanocytic nevi are all skin abnormalities that need to be worked on further before the model can consistently identify them.

Vakili et al. [5] describe a deep learning method to classify images of skin lesions such macules, nodules, papules, plaques, pustules, wheals, and bullas, with an eye on primary lesion classification and early-stage diagnosis. This framework employed deep learning techniques to categorize images of lesions into seven distinct groups. In this post, we experiment with several pre-trained deep convolutional neural networks to see which one produces the most accurate results. The results of the training and testing procedures showed that the ResNet-50 baseline model had improved to an acceptable 85.95% accuracy.

A state-of-the-art deep learning model for automatic multi-class classification of skin lesions was developed by Iqbal et al. [6]. Instead of using fewer filters and fewer parameters, the proposed Deep Convolutional Neural Network (DCNN) model takes use of additional layers and a more diverse range of filter sizes to improve efficiency and performance. Images acquired via dermoscopy are searched for in the ISIC-17, ISIC-18, and ISIC-19 databases maintained by the International Skin Imaging Collaboration. The aforementioned experimental results demonstrate the efficiency, reactivity, and specificity of the proposed DCNN approach. On the ISIC-17 accuracy scale, it scores 94%, while its sensitivity is 93%, and its specificity is 91%. The suggested DCNN method for skin lesion classification beats state-of-the-art algorithms, as shown by experimental findings, with an AUROC of 0.964 in the ISIC-17 dataset. This method may be utilized to assist dermatologists in this work. Considering the potential time, effort, and even human life savings that may follow from automating and speeding up the process of identifying skin lesions, the proposed technique is both novel and realistic.

Chaturvedi et al. [7] created a computer-aided diagnostic method for accurately classifying multi-class skin (MCS) cancers. The recommended strategy did better than both human professionals and the most effective deep learning algorithms when it came to identifying MCS tumors. Five pre-trained convolutional neural networks (CNNs) and four ensemble models were examined and improved upon using the HAM10000 dataset, with classification accuracy measured across seven classes. Our best single model had an accuracy of 93.20%, while our best ensemble model had an accuracy of 92.83%, according to this analysis. This framework recommends ResNeXt101 for MCS cancer classification because of its better design and greater accuracy.

Anjum et al. [8] proposed using convolutional neural networks (CNNs) for skin lesion detection. Important to the localization procedure is YOLOv2, which was created by layering ONNX with the Squeeze Net model. We conduct extensive tests to optimize the model's settings for precise lesion segmentation. The segmentation method achieves a Global Accuracy of 0.93 and 0.95 on ISBI 2017 and 2018, respectively. Classifying skin lesions using the ResNet-18 model and the cross entropy activation function. After collecting feature vectors, we utilize the ACO method to refine them. Classification accuracy was enhanced by the hybrid technique as compared to state-of-the-art methods.

An & et al. [9] proposed a transfer-learning strategy that makes use of an existing, trained Xception model. The original Xception model has one pooling layer, two thick layers, and a dropout layer added to it. After the previous FC layer was eliminated, seven skin disease categories were transferred to a new FC layer. The HAM10000 dataset was utilized to evaluate the proposed model because of its extreme skewness. Data augmentation techniques were used to complete the dataset and address the gaps in it. The latest results show that this model can accurately diagnose 96.40 percent of skin problems. When applied to the diagnosis of Benign Keratosis, the recommended model obtains a remarkable 99% accuracy, 97% sensitivity, and F1 score of 0.98. By reducing unnecessary anxiety and false alarms, this approach may provide patients and physicians a clearer picture of when medical intervention is required.

Srinivasu et al. [10] proposed using deep learning's Mobile Net V2 and Long Short-Term Memory (LSTM) to expedite the classification of skin diseases.

The Mobile Net V2 model is quick, precise, and resource-friendly. The proposed approach efficiently utilizes logs with refined data to provide trustworthy predictions. To monitor the transmission of disease, researchers use a co-occurrence matrix with shades of gray. Very Deep Convolutional Networks for Large-Scale Image Recognition (VGG) and an enlarged convolutional neural network architecture have also been used as performance benchmarks.

The article adopts a Convolutional Neural Network (CNN) constructed by Shanthi et al. [11], which is made up of numerous layers including the Convolution Layer, Activation Layer, Pooling Layer, Fully Connected Layer, and Soft-Max Classifier. Test cases from DermNet are used to evaluate the design. Acne, Keratosis, Eczema herpeticum, and Urticaria were the four main categories we pulled from the database, with 30-60 cases in each. Some of the challenges that must be overcome via automation include variations in skin tone, the location of sickness, and the need for specific equipment. The proposed CNN Classifier is accurate to within 0.04% of the actual classification rate.

Allugunti et al. [12] stated that a deep learning system may be able to effectively identify early melanoma subtypes. The three basic kinds of malignant lesions are identified in the proposed model: lesion malignant, superficial spreading, and nodular melanoma. This facilitates early diagnosis of the virus and the following isolation and treatment essential for eradicating it. Convolutional neural networks (CNNs) with deep layer topologies are used to demonstrate both DL and the more conventional non-parametric machine learning technique.

III. PROPOSED SYSTEM

Due to its recent rise in prominence in the area of image classification, CNN is used in this work to detect and categorize photos of skin disorders. The convolutional neural network was educated on data pertaining to a wide variety of skin lesions, such as actinic keratosis, basal cell carcinoma, dermatofibroma, melanoma, nevus, pigmented benign keratosis, squamous cell carcinoma, vascular lesion, and pigmented benign keratosis. Once trained, the CNN system can analyze any test picture and determine the disease shown. DL-CNN model-based CNN training:

Convolution layers such as a kernel or filter, rectified linear unit (ReLU), max pooling, fully connected layer, and SoftMax layer with classification layer are used to transform input images into probabilistic values in the range [0,1] for object classification during training and testing of a deep neural network or when employing transfer learning.

???????

Conclusion

In this research, we aimed to develop the most effective possible deep learning convolutional neural network (DL-CNN) for the multi-class classification of skin cancer. Therefore, the study\'s results might be used for the precise classification of all nine types of skin cancer. Future work will refine this study so that reinforcement learning may be effectively employed to properly categorize skin lesions.

References

[1] Zhang Ce, Xin Pan, Huapeng Li, Gardiner A, Sargent I, Jonathon S Hare, et al. A hybrid MLP-CNN classifier for very fine resolution remotely sensed image classification. Isprs Journal of Photogrammetry and Remote Sensing. 2017; 140:133-144. [2] Bi Lei, Jinman Kim, Euijoon Ahn, Feng D. Automatic Skin Lesion Analysis using Large-scale DermoscopyImages and Deep Residual Networks. ArXiv abs. 2017; 1703:04197. [3] N. Hameed, A. M. Shabut and M. A. Hossain, \"Multi-Class Skin Diseases Classification Using Deep Convolutional Neural Network and Support Vector Machine,\" 2018 12th International Conference on Software, Knowledge, Information Management & Applications (SKIMA), 2018, pp. 1-7, doi: 10.1109/SKIMA.2018.8631525. [4] Aldhyani, T.H.H.; Verma, A.; Al-Adhaileh, M.H.; Koundal, D. Multi-Class Skin Lesion Classification Using a Lightweight Dynamic Kernel Deep-Learning-Based Convolutional Neural Network. Diagnostics 2022, 12, 2048. https://doi.org/10.3390/diagnostics12092048 [5] N. Vakili, W. Krathu, and N. Laomaneerattanaporn. 2021. Multi-Class Primary Morphology Lesions Classification Using Deep Convolutional Neural Network. In The 12th International Conference on Advances in Information Technology (IAIT2021). Association for Computing Machinery, New York, NY, USA, Article 9, 1–7. https://doi.org/10.1145/3468784.3468887 [6] I. Iqbal, M. Younus, K. Walayat, M. U. Kakar, J. Ma, Automated multi-class classification of skin lesions through deep convolutional neural network with dermoscopic images, Computerized Medical Imaging and Graphics, Volume 88, 2021, 101843, ISSN0895-6111, https://doi.org/10.1016/j.compmedimag.2020.101843 [7] Chaturvedi, S.S., Tembhurne, J.V. & Diwan, T. A multi-class skin Cancer classification using deep convolutional neural networks. Multimed Tools Appl 79,28477–28498(2020). https://doi.org/10.1007/s11042-020-09388-2. [8] M. A. Anjum, J. Amin, M. Sharif, H. U. Khan, M. S. A. Malik and S. Kadry, \"Deep Semantic Segmentation and Multi-Class Skin Lesion Classification Based on Convolutional Neural Network,\" in IEEE Access, vol. 8, pp. 129668-129678,2020,doi:10.1109/ACCESS.2020.30 09276 [9] Anand, V., Gupta, S., Koundal, D., Nayak, S. R., Nayak, J., & Vimal, S. (2022). Multi-class Skin Disease Classification Using Transfer Learning Model. International Journal on Artificial Intelligence Tools, 31(02), 2250029 [10] Srinivasu PN, SivaSai JG, Ijaz MF, Bhoi AK, Kim W, Kang JJ. Classification of Skin Disease Using Deep Learning Neural Networks with MobileNet V2 and LSTM. Sensors. 2021; 21(8):2852. https://doi.org/10.3390/s21082852 [11] T. Shanthi, R.S. Sabeenian, R. Anand, Automatic diagnosis of skin diseases using convolution neural network, Microprocessors and Microsystems, Volume 76, 2020, 103074, ISSN 0141-9331, https://doi.org/10.1016/j.micpro.2020.103074 [12] Allugunti, V. R. (2022). A machine learning model for skin disease classification using convolutional neural networks. International Journal of Computing, Programming and Database Management, 3(1), 141-147.

Copyright

Copyright © 2024 P. V Venkateswa Rao, Nekkalapudi Gnanitha Chowdary, Chintha Sai Kiran, Guuduru Prerana, J. Sai Kiran, P. Dileep. This is an open access article distributed under the Creative Commons Attribution License, which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited.