Derin Konvolüsyon Ağı ile Dermatoskopik Görüntülerde Deri Lezyonlarının Sınıflandırılması

Bu çalışmada, cilt lezyonu içeren dermoskopik görüntülerin sınıflandırılması problemini çözmek için konvolüsyonel (evrişimsel) sinir ağı ile derin öğrenme temelli bir çözüm sunulmuştur. Klasik makine öğrenme teknikleri ile model tanımlamak hem çok zaman almakta hem de bu model ile veriyi ön işlem yapmadan anlamlı hale getirememektedir. Derin öğrenme sayesinde, uzun yıllar boyunca çözülmesinin zor olduğunu düşündüğümüz problemlerde büyük mesafe katedilmiştir. Derin öğrenme, elimizdeki veriyi bizim tarafımızdan herhangi bir müdahale olmadan kendisi işleyerek sonuca ulaşmaktadır. Deri lezyonunun dermoskopik görüntülerin sınıflandırılması melanositik tümörlerin iyi huylu veya kötü huylu olarak ayırt edilmesi zor bir görevdir. Malign melanom cilt kanserinin en ölümcül şeklidir ve dünyadaki en hızlı gelişen kanserlerden biridir. Erken teşhis edilirse kolayca tedavi edilebilir ve sonuçta melanomun erken teşhisi hayati öneme sahiptir. Dermoskopi melanom ve diğer pigmentli cilt lezyonlarının teşhisinde en önemli araçlardan biri haline gelmiştir. İnsan kararının yanlışlığı, öznelliği ve kötü tekrarlanabilirliği nedeniyle, dermoskopi görüntüsünün otomatik bir tanıma algoritması ile işlenmesi bir ihtiyaç olmuştur. 2017 yılında Muhammed Shadid ve Salman Khan tarafından yapılan uygulama ile 172 adet dermatoskopik görüntünün “benign” ve “malignant” olarak iki sınıfa ayrıştırılmasında Support Vector Machine(SVM) sınıflandırıcısı kullanılmıştır. Veri kümesi üzerinde yapılan deneyler %91 doğruluk hassasiyetine sahiptir. Ancak bizim veri setimizde binlerce imge olması ve yedi adet lezyon sınıfına ayrıştıracak olmamız daha etkin yöntemler aramamızı gerektirmiştir. Melanomların sınıf içi tutarsızlığı, cilt lezyonlarının düşük kontrastı ve dermoskopi görüntülerinde gürültü, saçın varlığı, hava kabarcıkları ve melanom olmayan vakalar arasındaki benzerlik gibi yapay nesneler nedeniyle zorlu bir süreç olarak kabul edilir. Bu problemleri çözmek için, dermoskopik görüntülerde farklı yedi hastalık tipini sınıflandırmak için güçlü bir evrişimsel sinir ağı modeli içeren VGGNET-16 mimarisini önermekteyiz. “HAM10000” (Human Againist Machine) veri seti üzerinde VGGNET-16 mimarisi ile tasarlanan derin ağ modeli eğitilerek sonuçlar gözlemlenmiştir.Akademik makine öğrenmesi için bir eğitim seti olarak yayınlanan ve ISIC arşivi aracılığıyla kamuya açık olan veri seti 10015 adet dermatoskopik görüntüden oluşmaktadır. Toplam yedi adet lezyon sınıfının bulunduğu veri setinin eğitim ve test alanı olarak ayrıştırılmasında K-Fold Cross Validation tekniğinden faydalanılmıştır. Eğitilmiş modelin test aşamasında sınıfların onaylama doğruluğu%85,62olarak elde edilmiştir.

Classification of Skin Lesions in Dermatoscopic Images with Deep Convolution Network

In this study, a deep learning based solution with convolutional neural network is presented to solve the problem of classificationof dermoscopic images containing skin lesions. Defining models with classical machine learning techniques takes a lot of timeand with this model, it cannot make data meaningful without pretreatment. Thanks to deep learning, we have come a long wayin problems that we think are difficult to solve for many years. Deep learning achieves results by processing the data at handwithout any intervention by us.Classification of dermoscopic images of skin lesions is a difficult task to distinguish between benign or malignant melanocytictumors. Malignant melanoma is the deadliest form of skin cancer and is one of the fastest developing cancers in the world. Ifdiagnosed early, it can be easily treated and ultimately, early diagnosis of melanoma is vital. Dermoscopy has become one of themost important tools in the diagnosis of melanoma and other pigmented skin lesions. Due to the inaccuracy, subjectivity, andpoor reproducibility of human judgment, there has been a need to process the dermoscopy image with an automatic recognitionalgorithm.In 2017, the support vector machine (SVM) classifier was used to differentiate 172 dermatoscopic images into two classes as“benign”and “malignant”. Experiments on the dataset have 91% accuracy. However, the fact that we have thousands of imagesin our data set and that we will break them down into seven lesion classes required us to search for more effective methods.Classroom inconsistency of melanomas is considered a challenging process due to the low contrast of skin lesions and artificialobjects such as noise, presence of hair, air bubbles and similarity between non-melanoma cases in dermoscopy images. To solvethese problems, we propose the VGGNET-16 architecture, which includes a powerful convolutional neural network model toclassify seven different types of disease on dermoscopic images.“HAM10000 ”(Human Againist Machine) data set was used with VGGNET-16 architecture and the results were observed. Thedata set, which is published as an educational set for academic machine learning and made public through the ISIC archive,consists of 10015 dermatoscopic images. K-Fold Cross Validation technique was used to differentiate the data set consisting ofseven lesion classes as training and test area. In the test phase of the educated model, the validation of the classes was obtainedas 85.62%.

___

  • [1] Schmidhuber, J. (2015). Deep learning in neural networks: An overview. Neural networks, 61, 85-117.
  • [2] LeCun, Y., Bengio, Y., & Hinton, G. (2015). Deep learning. nature, 521(7553), 436.
  • [3] Bottou, L. (2010). Large-scale machine learning with stochastic gradient descent. In Proceedings of COMPSTAT'2010(pp. 177-186). Physica-Verlag HD.
  • [4] Simard, P. Y., Steinkraus, D., & Platt, J. C. (2003, August). Best practices for convolutional neural networks applied to visual document analysis. In Icdar(Vol. 3, No. 2003).
  • [5] Deng, L.,& Yu, D. (2014). Deep learning: methods and applications. Foundations and Trends® in Signal Processing, 7(3–4), 197- 387.
  • [6] Ozgur, A. (2004). Supervised and unsupervised machine learning techniques for text document categorization. Unpublished Master’s Thesis, İstanbul: Boğaziçi University.
  • [7] Song, H. A.,& Lee, S. Y. (2013, November). Hierarchical representation using NMF. In International conference on neural information processing(pp. 466-473). Springer, Berlin, Heidelberg.
  • [8] Ivakhnenko, A. G.,& Lapa, V. G. (1966). Cybernetic predicting devices (No. TR-EE66-5). PURDUE UNIV LAFAYETTE IND SCHOOL OF ELECTRICAL ENGINEERING.
  • [9] Dettmers, T. (2015). Deep learning in a nutshell: Core concepts. NVIDIA Devblogs.
  • [10] Fukushima, K. (1980). Neocognitron: A self-organizing neural network model for a mechanism of pattern recognition unaffected by shift in position. Biological cybernetics, 36(4), 193-202.
  • [11] LeCun, Y., Boser, B., Denker, J. S., Henderson, D., Howard, R. E., Hubbard, W., & Jackel, L. D. (1989). Backpropagation applied to handwritten zip code recognition. Neural computation, 1(4), 541-551.
  • [12] LeCun, Y., Boser, B. E., Denker, J. S., Henderson, D., Howard, R. E., Hubbard, W. E., & Jackel, L. D. (1990). Handwritten digit recognition with a back-propagation network. In Advances in neural information processing systems(pp. 396-404).
  • [13]Hinton, G. E., Dayan, P., Frey, B. J., & Neal, R. M. (1995). The" wake-sleep" algorithm for unsupervised neural networks. Science, 268(5214), 1158-1161.
  • [14]Hochreiter, S.,& Schmidhuber, J. (1997). Long short-term memory. Neural computation, 9(8), 1735-1780.
  • [15]Aizenberg, I. N., Aizenberg, N. N., & Vandewalle, J. (2000). Multiple-Valued Threshold Logic and Multi-Valued Neurons. In MultiValued and Universal Binary Neurons(pp. 25-80). Springer, Boston, MA.
  • [16]Hinton, G. E. (2007). Learning multiple layers of representation. Trends in cognitive sciences, 11(10), 428-434.
  • [17] Lawrence, S., Giles, C. L., Tsoi, A. C., & Back, A. D. (1997). Face recognition: A convolutional neural-network approach. IEEE transactions on neural networks, 8(1), 98-113.
  • [18] Hijazi, S., Kumar, R., & Rowen, C. (2015). Using convolutional neural networks for image recognition. Cadence Design Systems Inc.: San Jose, CA, USA.
  • [19]Le Cun, Y., Jackel, L. D., Boser, B., Denker, J. S., Graf, H. P., Guyon, I., ... & Hubbard, W. (1989). Handwritten digit recognition: Applications of neural network chips and automatic learning. IEEE Communications Magazine, 27(11), 41-46.
  • [20]F. Vazquez, “ https://towardsdatascience.com/a-weird-introduction-to-deep-learning-7828803693b0”
  • [21] Binder, M., Steiner, A., Schwarz, M., Knollmayer, S., Wolff, K., & Pehamberger, H. (1994). Application of an artificial neural network in epiluminescence microscopy pattern analysis of pigmented skin lesions: a pilot study. British Journal of Dermatology, 130(4), 460-465.
  • [22] Mendonça, T., Ferreira, P. M., Marques, J. S., Marcal, A. R., & Rozeira, J. (2013, July). PH 2-A dermoscopic image database for research and benchmarking. In 2013 35th annual international conference of the IEEE engineering in medicine and biology society (EMBC)(pp. 5437-5440). IEEE.
  • [23] Tschandl, P., Rosendahl, C., & Kittler, H. (2018). The HAM10000 dataset, a large collection of multi-source dermatoscopic images of common pigmented skin lesions. Scientific data, 5, 180161.
  • [24] Reddy, N. D. (2018). Classification of Dermoscopy Images using Deep Learning. arXiv preprint arXiv:1808.01607.
  • [25] Khan, M. A., Javed, M. Y., Sharif, M., Saba, T., & Rehman, A. (2019, April). Multi-Model Deep Neural Network based Features Extraction and Optimal Selection Approach for Skin Lesion Classification. In 2019 International Conference on Computer and Information Sciences (ICCIS)(pp. 1-7). IEEE.