Skin lesion classification using convolutional neural networks based on Multi-Features Extraction

CAIP 2021

Transcript

1. Skin lesion classification using convolutional neural networks based on Multi-Features Extraction Samia Benyahia 1, Boudjelal Meftah 2 , Olivier Lezoray 3 The 19th International Conference on Computer Analysis of Images and Patterns , September 2021 1 Department of Computer Science, Faculty of Exact Sciences, University of Mascara, Mascara, Algeria 2 LRSBG Laboratory, University of Mascara, Mascara, Algeria 3 Normandie Univ, UNICAEN, ENSICAEN, CNRS, GREYC, Caen, France
2. Introduction • Cancer is a generic term for a large group of diseases that can affect many different cell types and organs of the body • Arises from the transformation of normal cells into irregular cells, characterized by the uncontrolled growth and spread of abnormal cells that grow beyond their usual boundaries, results in a mass (tumor) • Metastases can result in death if not treated • 1.9 million new cancer cases are expected to be diagnosed in 2021 2
3. Skin cancer : Melanoma 3 • Skin cancer is one of the deadly types of cancer that has spread worldwide • Classes of skin lesions : Melanoma (MEL), Melanocytic nevus (NV), Basal cell carcinoma (BCC), Actinic Keratosis (AK), Benign keratosis lesion, Dermatofibroma (DF), Vascular lesion (VASC), and Squamous cell carcinoma (SCC) • Melanoma is a cancer of pigment producing-cells called melanocytes • Melanoma is rare and represents only 5% of all the cancer types, but it is held responsible for more than 70% of the mortality affected
4. Skin cancer 4 • The classification of the skin lesion plays an important role in the early detection and diagnosis of skin cancer • Wrong recognition are relatively high because of the high similarity among different lesion classes • Automatic classification of melanoma is required to have enhanced accuracy and enough efficiency to detect the skin lesion • Handcrafted features extraction : statistical pixel-level features, shape features, texture features, and relational features • Deep-learning technique is far better than the traditional methods in particular Convolutional Neural Networks
5. Methodology 5 Proposed method's fowchart
6. PH2 database • 200 dermoscopic images • 80 common nevus • 80 atypical nevus • 40 melanomas 6
7. Convolutional neural networks 7 • Convolutional, maxpool, batch normalization, activation, and a residual blocks, fully connected layers • Skip-connection between layers • ResNet18, ResNet34, ResNet50, ResNet101, and ResNet152 • Convolution, pooling layers, transition layers, dense blocks, fully connected layers • Dense blocks • DensNet-121, DensNet-169,DensNet- 201, and DensNet-246 • Scaling the different dimensions such as depth, width, and image resolution of the network at the same time • Family of eight different CNN models: EcientNet-B0 to B7 ResNet DenseNet EfficientNet
8. Classifiers • Artificial Neural Network (ANN) • Support Vector Machines (SVMs) • K-Nearest Neighbor (KNN) • Random Forest (RF) 8
9. Results • Classification of skin lesions into two types of lesions: melanoma or non- melanoma • Classification of skin lesions into three types: melanoma, atypical nevus, or common nevus 9 80% 85% 90% 95% 100% DenseNet201 ResNet50 EffcientB0 SVM KNN ANN RF Accuracy according to DenseNet201, ResnNet50, and EfficientNetB0 with four classiers for melanoma or non-melanoma. 0,00% 20,00% 40,00% 60,00% 80,00% DenseNet201 ResNet50 EffcientB0 SVM KNN ANN RF Accuracy according to DenseNet201, ResnNet50, and EfficientNetB0 with four classifiers for melanoma, atypical nevus or common nevus without augmentation. Accuracy according to DenseNet201, ResnNet50, and EfficientNetB0 with four classifiers for melanoma, atypical nevus or common nevus with augmentation. 60% 65% 70% 75% 80% 85% 90% 95% 100% DenseNet201 ResNet50 EffcientB0 SVM KNN ANN RF Densenet201-SVM and RF acheived accuracy = 99% Densenet201-KNN-ANN acheived accuracy = 70% Resenet50-EfficientB0-RF acheived accuracy = 70% Densenet201-SVM acheived accuracy = 95% Resenet50-ANN acheived accuracy = 95%
10. Results 10 Authors Method Accuracy Ghasem[2] SBS 98.50% Ghasem[2] HSBS 96.70% Ann[3] AlexNet 93.00% Filali[5] SVM 98.00% Sanket[6] SVM 92.00% Proposed DenseNet201+SVM 99% Authors Method Accuracy Ozkan[1] MLP 92.50% Ann[3] AlexNet 67.50% Singh[4] SVM 92.50% Khalid[7] AlexNet 98.61% Proposed DenseNet+SVM 95% Results of the proposed approach compared to various approaches for PH2 for melanoma and non- melanoma. Results of the proposed approach compared to various approaches for PH2 dataset for melanoma, atypical nevus, and common nevus.
11. Conclusion • Three pre-trained CNN architectures as feature extractors and four classifiers to evaluate the classification of skin lesions from PH2 datasets with two or three classes • DenseNet201 combined with SVM classifiers achieved very good accuracy of 99% for melanoma and non-melanoma detection while 95% for melanoma, atypical nevus, or common nevus • Large sample size dataset, metadata, ensemble method 11
12. References 1. Ozkan, I., Koklu, M.: Skin Lesion Classication using Machine Learning Algorithms. International Journal of Intelligent Systems and Applications in Engineering 5, 285-289 (2017) 2. Ghasem Shakourian, G., Kordy, H. M., Ebrahimi, F.: A hierarchical structure based on Stacking approach for skin lesion classication. Expert Syst. Appl. 145, 113- 127 (2020) 3. Salido, Julie Ann A., Ruiz, C. R.: Using Deep Learning to Detect Melanoma in Dermoscopy Images. International Journal of Machine Learning and Computing 8(1), 61-68 (2018) 4. Singh, L., Janghel, R. R., Sahu, S.: Designing a Retrieval-Based Diagnostic Aid using Eective Features to Classify Skin Lesion in Dermoscopic Images. Procedia Computer Science 167 : 2172{2180 (2020) 5. Filali, Y., El Khoukhi, H., Sabri, M., Aarab, A.: Ecient fusion of handcrafted and pre-trained CNNs features to classify melanoma skin cancer. Multimedia Tools and Applications, 1-20 (2020) 6. Sanket, K., Chandra, J.: Skin Cancer Classication using Machine Learning for Dermoscopy Image 1457 (2019) 7. Khalid, M. H., Kassem, M. A., Foaud, M. M.: Skin Cancer Classication using Deep Learning and Transfer Learning. 2018 9th Cairo 8. International Biomedical Engineering Conference (CIBEC), 90{93 (2018) 9. Howlader N, Noone AM, Krapcho M, et al. (eds). SEER Cancer Statistics Review, 1975–2018. National Cancer Institute, posted to the SEER website, April 2021. Last accessed April 19, 2021 10. He, K., Zhang, X., Ren, S., Sun, J.:Deep Residual Learning for Image Recognition,IEEE Conference on Computer Vision and Pattern Recognition (CVPR), 770- 778(2016) 11. Huang, G., Liu, Z.,Weinberger, K. Q.:Densely Connected Convolutional Networks, IEEE Conference on Computer Vision and Pattern Recognition (CVPR), 2261- 2269 (2017) 12. Tan, M., Le, Q.V.: EcientNet: Rethinking Model Scaling for Convolutional Neural Networks, ArXiv (2019) 13. Mendonan, T. and Ferreira, P. and Marques, J. and Maral, A. and Rozeira, J.:PH2 - A dermoscopic image database for research and benchmarking, 35th Annual International Conference of the IEEE Engineering in Medicine and Biology Society (EMBC), 5437-5440 (2013) 12
13. Thank you for your attention 13