Skin cancer detection using Deep learning architecture-Deep Learning Project- Matlab
Skin cancer detection using Deep learning architecture-Deep Learning Project
Platform : Matlab Delivery : One Working Day Support : Online Demo ( 2 Hours)
99 in stock
The mortality rate of skin cancer is very high, but it can usually be cured by minor surgery. A Fast and accurate diagnosis can greatly benefit doctors and patients. In many medical fields, the performance of the recent deep-learning-based model is close to or even exceeds the level of human experts. In this paper, to help dermatologists improve the efficiency of skin cancer analysis, we employ the DenseNet model to complete the recognition of skin lesion images. The proposed model is trained and evaluated with the ISIC2020 dataset. Besides, Experimental results show that our method achieves superior performance over the other deep-learning approaches.
Skin cancer is one of the most common cancers. Melanoma, which has a low incidence of skin cancer, has a mortality rate of nearly 75%. The American Cancer Society estimates that more than 100,000 new melanoma cases will be diagnosed by 2020, and nearly 7,000 people are expected to die from the disease. Generally speaking, early detection based on data science can help us improve the efficiency of treatment. Furthermore, the use of image diagnostic tools that can intelligently identify melanoma can also improve the diagnostic accuracy of dermatologists, which will have the opportunity to positively affect millions of people and make a huge difference in the field of medicine. Therefore, the research topic in this paper is to solve the SIIM-ISIC melanoma classification challenge provided by the Kaggle competition. In this competition, we will propose an effective method to identify melanoma in images of skin lesions. Usually, a dermatologist evaluates each mole of a patient to find outliers or “ugly ducklings” that are most likely to be melanoma. However, the existing AI methods have not yet fully considered this clinical reference frame. Thus in this article, we aim to use contextual information at the patient level to enhance the accuracy of melanoma recognition, so as to better support clinical dermatologists and improve their diagnostic accuracy and efficiency. In this paper, we conduct experiments on the Kaggle competition dataset (ISIC2020) which contains the largest publicly available collection of quality-controlled dermoscopic images of skin lesions. We propose to use the DenseNet model to train and evaluate the ISIC2020 dataset provided in this competition. At the same time, we also compared the effect of this model with the VGG and ResNet networks. The experimental results show that our DenseNet model can finally achieve 0.925 AUC on the test data set, which is 0.034 higher than VGG and 0.003 higher than ResNet. In summary, the contributions of this article are as follows: • We propose to use the DenseNet201 model pre-trained on imagenet to train and evaluate the skin lesion image data, which can take good advantage of patient-level contextual information. • We evaluate our DenseNet method on the ISIC2020 dataset. Experiments show that our method outperforms the comparison approaches, like VGG and ResNet model (0.034 higher than VGG and 0.003 higher than ResNet with AUC metric, and finally gain 0.925AUC).
Once melanoma is diagnosed, it can be cured by surgical removal as early as possible. Nevertheless, accurate diagnosis of melanoma is special and needs expert knowledge. However, the prognosis of more advanced cases is often limited. In clinical routine, the sensitivity of the detection of melanoma is very important; meanwhile, we should avoid the misjudgment of benign nevus. In this section, we mainly introduce the recent development of melanoma recognition and the deep-learning-based backbone network including DenseNet . Therefore, the rest of this article is organized as follows. The second section introduces the understanding of melanoma. Next, our model DenseNet201 is introduced in detail in the third part. Then, in Section 4, we conducted experiments between our proposed method and other comparison models through the AUC metric. Finally, the fifth section describes the conclusions of this article and future work.
- The proposed system use a deep learning based CNN architecture.
- Where the system improve the accuracy based on learning
- To improve the accuracy in terms of specificity and sensitivity.
In this paper, we propose to use DenseNet to realize the recognition of melanoma in skin lesion pictures. We use the imagenet pre-trained DenseNet201 model on the ISIC2020 data set provided on the Kaggle competition platform. Experiments show that compared with VGG16 and ResNet50 networks, this model can more effectively identify melanoma and help clinical dermatologists improve the accuracy of the diagnosis of melanoma. In the future, we will improve the model to improve the prediction effect and apply the model to other related medical fields.
 Huang, G., Liu, Z., & Weinberger, K.Q. (2017). Densely Connected Convolutional Networks. 2017 IEEE Conference on Computer Vision and Pattern Recognition (CVPR), 2261-2269.
 Bafounta, M., Beauchet, A., Aegerter, P, & Saiag, P (2001). Is dermoscopy (epiluminescence microscopy) useful for the diagnosis of melanoma? Results of a meta-analysis using techniques adapted to the evaluation of diagnostic tests. Archives of dermatology, 137 10, 1343-50 .
 Kittler, H., Pehamberger, H., Wolff, K., & Binder, M. (2002). Diagnostic accuracy of dermoscopy. The Lancet. Oncology, 3 3, 159-65 .
 Vestergaard, M., Macaskill, P, Holt, PE., & Menzies, S. (2008). Dermoscopy compared with naked eye examination for the diagnosis of primary melanoma: a meta – analysis of studies performed in a clinical setting. British Journal of Dermatology, 159.
 Pehamberger, H., Steiner, A., & Wolff, K. (1987). In vivo epiluminescence microscopy of pigmented skin lesions. I. Pattern analysis of pigmented skin lesions. Journal of the American Academy of Dermatology, 17 4, 571-83