SegNet Mimarisi Kullanılarak Cilt Lezyon Bölütleme Performansının İyileştirilmesi

Öz

Kötü huylu melanom, bütün cilt kanseri türleri arasında üçüncü en sık rastlanan tür olmasına rağmen en çok ölüme neden olan formudur. Kötü huylu melanomun erken aşamada teşhisi hastanın yaşama şanşını büyük oranda artırdığından, erken teşhis oldukça önemlidir. Melanom teşhisi dermatologlar tarafından lezyon bölgesinin geometrisi, rengi, yapısal ve dokusal özellikleri gibi görsel niteliklerine bakılarak yapılmaktadır. Ancak, son zamanlarda, bilgisayarlı görü ve makine öğrenmesi yöntemlerindeki gelişmeler ile birlikte melanoma tanısı için bilgisayar destekli tanı sistemleri popüler olmaya başlamıştır. Bu çalışmada ciltte bulunan lezyonların bölütlenmesi için SegNet mimarisi tabanlı bir sistem geliştirilmiştir. Bunun yanında, cilt lezyonları üzerinde veri büyütme ve renk tutarlılığı ve kıl silme gibi önişleme adımların bölütleme performansı üzerinde etkileri incelenmiştir. Deneylerimizde ISBI2016 veri kümesi kullanılmıştır. Sonuçlar veri büyütme ve önişlemenin bölütleme performasını dikkate değer oranda artırdığını göstermektedir. Bununla birlikte, veri büyütmenin ezberlemeyi önlediği ve modellerin genelleme yeteneğini artırdığı sonucuna varılmıştır.

Anahtar Kelimeler

• melanoma,bölütleme,derin öğrenme,lezyon,segnet

Improving Skin lesion Segmentation Performance Using SegNet Architecture

Abstract

The malignant melanoma is the third most common form of skin cancer among all skin cancer types, but it is the most fatal form of skin cancer. Early diagnosis is very important, as the early diagnosis of malignant melanoma greatly increases the patient's survival chance. The melanoma diagnosis is carried out by dermatologists by examining the visual characteristics of the lesion area such as geometry, color, structural and textural features. However, recently, the computer-aided diagnosis systems have become popular for the melanoma detection, with advances in computer vision and machine learning methods. In this study, a SegNet architecture based system has been developed for segmentation of skin lesions. In addition, the effects of preprocessing steps on skin lesions such as data augmentation and color consistency and hair removal were investigated on segmentation performance. ISBI2016 dataset was used in our experiments. The results show that data augmentation and preprocessing significantly increases segmentation performance. However, it was concluded that data augmentation prevents memorization and increases the generalization ability of the models.

Keywords

• Melanoma,Segmentation,Deep Learning,Skin Lesion,SegNet

Kaynakça

1. Agarwal, A., Issac, A., Dutta, M. K., Riha, K. ve Uher, V. (2017). Automated skin lesion segmentation using k-Means clustering from digital dermoscopic images. 2017 40th International Conference on Telecommunications and Signal Processing, TSP 2017 içinde (C. 2017-January, ss. 743–748). Institute of Electrical and Electronics Engineers Inc. doi:10.1109/TSP.2017.8076087
2. Ahn, E., Bi, L., Jung, Y. H., Kim, J., Li, C., Fulham, M. ve Feng, D. D. (2015). Automated saliency-based lesion segmentation in dermoscopic images. Proceedings of the Annual International Conference of the IEEE Engineering in Medicine and Biology Society, EMBS, 2015-Novem, 3009–3012. doi:10.1109/EMBC.2015.7319025
3. Ahn, E., Kim, J., Bi, L., Kumar, A., Li, C., Fulham, M. ve Feng, D. D. (2017). Saliency-Based Lesion Segmentation Via Background Detection in Dermoscopic Images. IEEE Journal of Biomedical and Health Informatics, 21(6), 1685–1693. doi:10.1109/JBHI.2017.2653179
4. Argenziano, G., Fabbrocini, G., Carli, P., De Giorgi, V., Sammarco, E. ve Delfino, M. (1998). Epiluminescence microscopy for the diagnosis of doubtful melanocytic skin lesions: Comparison of the ABCD rule of dermatoscopy and a new 7-point checklist based on pattern analysis. Archives of Dermatology, 134(12), 1563–1570. doi:10.1001/archderm.134.12.1563
5. Argenziano, G. ve Soyer, H. P. (2001). Dermoscopy of pigmented skin lesions - a valuable tool for early diagnosis of melanoma. Lancet Oncology, 2(7), 443–449. doi:10.1016/S1470-2045(00)00422-8
6. Badrinarayanan, V., Kendall, A. ve Cipolla, R. (2017). SegNet: A Deep Convolutional Encoder-Decoder Architecture for Image Segmentation. IEEE Transactions on Pattern Analysis and Machine Intelligence, 39(12), 2481–2495. doi:10.1109/TPAMI.2016.2644615
7. Barata, C., Celebi, M. E. ve Marques, J. S. (2015). Improving dermoscopy image classification using color constancy. IEEE Journal of Biomedical and Health Informatics, 19(3), 1146–1152. doi:10.1109/JBHI.2014.2336473
8. Bi, L., Kim, J., Ahn, E., Feng, D. ve Fulham, M. (2017). Semi-automatic skin lesion segmentation via fully convolutional networks. Proceedings - International Symposium on Biomedical Imaging, 561–564. doi:10.1109/ISBI.2017.7950583

9. Bi, L., Kim, J., Ahn, E., Feng, D., Fulham, M., Medicine, N., … Hospital, A. (2016). Automated Skin Lesion Segmentation via Image-wise Supervised Learning and Multi-Scale Superpixel Based Cellular Automata School of Information Technologies , University of Sydney , Australia Sydney Medical School , University of Sydney , Australia Med-X R, 1059–1062.
10. Brahmbhatt, P. ve Rajan, S. N. (2019). Skin Lesion Segmentation using SegNet with Binary Cross- Entropy, 14–15.
11. Burg, G. (1993). Das Melanom (Serie Gesundheit) . Piper/VCH.
12. Chan, T. F. ve Vese, L. A. (2001). Active contours without edges. IEEE Transactions on Image Processing, 10(2), 266–277. doi:10.1109/83.902291
13. Erkol, B., Moss, R. H., Stanley, R. J., Stoecker, W. V ve Hvatum, E. (2005). Images Using Gradient Vector Flow Snakes. Skin Research and Technology, 17–26.
14. Garnavi, R., Aldeen, M., Celebi, M. E., Varigos, G. ve Finch, S. (2011). Border detection in dermoscopy images using hybrid thresholding on optimized color channels. Computerized Medical Imaging and Graphics, 35(2), 105–115. doi:10.1016/j.compmedimag.2010.08.001
15. Kasmi, R., Mokrani, K., Rader, R. K., Cole, J. G. ve Stoecker, W. V. (2016). Biologically inspired skin lesion segmentation using a geodesic active contour technique. Skin Research and Technology, 22(2), 208–222. doi:10.1111/srt.12252
16. L.A.G. Ries, D. M. et al. (2008). SEER Cancer Statistics Review 1975-2005 . Bethesda, MD, National Cancer Institute. 13 Mart 2020 tarihinde https://seer.cancer.gov/archive/csr/1975_2005/ adresinden erişildi.
17. Lee, T., Ng, V., Gallagher, R., Coldman, A. ve McLean, D. (1997). Dullrazor®: A software approach to hair removal from images. Computers in Biology and Medicine, 27(6), 533–543. doi:10.1016/S0010-4825(97)00020-6
18. Menzies, S. W., Ingvar, C., Crotty, K. A. ve McCarthy, W. H. (1996). Frequency and morphologic characteristics of invasive melanomas lacking specific surface microscopic features. Archives of Dermatology, 132(10), 1178–1182. doi:10.1001/archderm.132.10.1178
19. Romero Lopez, A., Giro-I-Nieto, X., Burdick, J. ve Marques, O. (2017). Skin lesion classification from dermoscopic images using deep learning techniques. Proceedings of the 13th IASTED International Conference on Biomedical Engineering, BioMed 2017, 49–54. doi:10.2316/P.2017.852-053
20. Schmid-Saugeon, P., Guillod, J. ve Thiran, J. P. (2003). Towards a computer-aided diagnosis system for pigmented skin lesions. Computerized Medical Imaging and Graphics, 27(1), 65–78. doi:10.1016/S0895-6111(02)00048-4
21. Stolz, W., Reimann, A. ve Cognetta, A. B. (1994). ABCD rule of dermatoscopy: a new practical method for early recognition of malignant melanoma.
22. Tang, P., Liang, Q., Yan, X., Xiang, S., Sun, W., Zhang, D. ve Coppola, G. (2019). Efficient skin lesion segmentation using separable-Unet with stochastic weight averaging. Computer Methods and Programs in Biomedicine, 178, 289–301. doi:10.1016/j.cmpb.2019.07.005
23. Wong, A., Scharcanski, J. ve Fieguth, P. (2011). Automatic skin lesion segmentation via iterative stochastic region merging. IEEE Transactions on Information Technology in Biomedicine, 15(6), 929–936. doi:10.1109/TITB.2011.2157829
24. Xie, F. ve Bovik, A. C. (2013). Automatic segmentation of dermoscopy images using self-generating neural networks seeded by genetic algorithm. Pattern Recognition, 46(3), 1012–1019. doi:10.1016/j.patcog.2012.08.012