Neural Network Based Classification of Melanocytic Lesions in Dermoscopy: Role of Input Vector Encoding

Year 2018, Volume: 18 Issue: 2, 242 - 248, 03.08.2018

Gökhan Ertaş

Abstract

DOI: 10.26650/electrica.2018.97856

Melanocytic lesions are the main cause of
death from skin cancer, and early diagnosis is the key to decreasing the
mortality rate. This study assesses the role of input-vector encoding in neural
network-based classification of melanocytic lesions in dermoscopy. Twelve
dermoscopic measures from 200 melanocytic lesions are encoded by compact
encoding, adrenocortical dysplasia (ACD) encoding, 1-of-N encoding, normalized
encoding, and raw encoding, resulting in five different input-vector sets.
Feed-forward neural networks with one hidden layer and one output layer are
designed with several neurons in the hidden layer, ranging from two to
twenty-two for each type of input-vector set, to classify a melanocytic lesion
into common nevus, atypical nevus, and melanoma. Accordingly, 105 networks are
designed and trained using supervised learning and then tested by performing a 10-fold
cross validation. All the neural networks achieve high sensitivities,
specificities, and accuracies in classification. However, the network with
seven neurons in the hidden layer and raw encoded dermoscopic measures as the
input vector realizes the highest sensitivity (97.0%), specificity (98.1%), and
accuracy (98.0%). The practical use of the network can facilitate lesion
classification by retaining the needed expertise and minimizing diagnostic
variability among dermatologists.

Keywords

Dermoscopy, melanocytic lesion, multilayer perceptron, encoding, classification, classification

Neural Network Based Classification of Melanocytic Lesions in Dermoscopy: Role of Input Vector Encoding

Abstract

DOI: 10.26650/electrica.2018.97856

Melanocytic lesions are the main cause of death from skin cancer, and early diagnosis is the key to decreasing the mortality rate. This study assesses the role of input-vector encoding in neural network-based classification of melanocytic lesions in dermoscopy. Twelve dermoscopic measures from 200 melanocytic lesions are encoded by compact encoding, adrenocortical dysplasia (ACD) encoding, 1-of-N encoding, normalized encoding, and raw encoding, resulting in five different input-vector sets. Feed-forward neural networks with one hidden layer and one output layer are designed with several neurons in the hidden layer, ranging from two to twenty-two for each type of input-vector set, to classify a melanocytic lesion into common nevus, atypical nevus, and melanoma. Accordingly, 105 networks are designed and trained using supervised learning and then tested by performing a 10-fold cross validation. All the neural networks achieve high sensitivities, specificities, and accuracies in classification. However, the network with seven neurons in the hidden layer and raw encoded dermoscopic measures as the input vector realizes the highest sensitivity (97.0%), specificity (98.1%), and accuracy (98.0%). The practical use of the network can facilitate lesion classification by retaining the needed expertise and minimizing diagnostic variability among dermatologists.

Keywords

Dermoscopy, melanocytic lesion, multilayer perceptron, encoding, classification

References

• 1. P. Corrie, M. Hategan, K. Fife, C. Parkinson, “Management of melanoma”, Br Med Bull, vol. 111, no. 1, pp. 149-162, 2014.
• 2. C. Cerchia, A. Lavecchia, “Small molecule drugs and targeted therapy for melanoma: Current strategies and future directions”, Curr Med Chem, vol. 24, pp. 2312-2344, 2017.
• 3. V. Gray-Schopfer, C. Wellbrock, R. Marais, “Melanoma biology and new targeted therapy”, Nature, vol. 445, no. 7130, pp. 851-857, 2007.
• 4. V. P. Silva, J. K. Ikino, M. M. Sens, D. H. Nunes, G. Di Giunta, “Dermoscopic features of thin melanomas: a comparative study of melanoma in situ and invasive melanomas smaller than or equal to 1mm”, An Bras Dermatol, vol. 88, no. 5, pp. 712-717, 2013.
• 5. R. J. Friedman, D. S. Rigel, A. W. Kopf, “Early detection of malignant melanoma: the role of physician examination and self-examination of the skin”, CA: Cancer J Clin, vol. 35, no. 3, pp. 130-151, 1985.
• 6. S. W. Menzies, C. Ingvar, K. A. Crotty, W. H. McCarthy, “Frequency and morphologic characteristics of invasive melanomas lacking specific surface microscopic features”, Arch Dermatol, vol.132, no. 10, pp. 1178-1182, 1996.
• 7. G. Argenziano, G. Fabbrocini, P. Carli, V. De Giorgi, E. Sammarco, M. Delfino, “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”, Arch Dermatol, vol. 134, pp. 1563-1570, 1998.
• 8. C. Carrera, M. A. Marchetti S. W. Dusza, G. Argenziano, R. P. Braun, A. C. Halpern, N. Jaimes, H. J. Kittler, J. Malvehy, S. W. Menzies, G. Pellacani, S. Puig, H. S. Rabinovitz, A. Scope, H. P. Soyer, W. Stolz, R. Hofmann-Wellenhof, I. Zalaudek, A. A. Marghoob, “Validity and reliability of dermoscopic criteria used to differentiate nevi from melanoma: A web-based international dermoscopy society study”, JAMA Dermatol, vol. 152, no. 7, pp 798-806, 2016.
• 9. U. Fidan, I. Sari, R. K. Kumrular, “Classification of skin lesions using ANN”, Medical Technologies National Congress (TIPTEKNO), 2016, pp. 1-4.
• 10. A. Bastürk, M. E. Yuksel, H. Badem, A. Calışkan, “Deep neural network based diagnosis system for melanoma skin cancer”, Signal Processing and Communications Applications Conference (SIU), 2017, pp. 1-4.
• 11. I. A. Özkan, M. Koklu, “Skin Lesion Classification using Machine Learning Algorithms”, International Journal of Intelligent Systems and Applications in Engineering, vol. 5, no. 4, pp 285-289, 2017.
• 12. T. Mendonça, P. M. Ferreira, J. S. Marques, A. R. Marcal, J. Rozeira, “PH 2-A dermoscopic image database for research and benchmarking”, Engineering in Medicine and Biology Society (EMBC), 35th Annual International Conference of the IEEE, 2013, pp. 5437-5440.
• 13. T. T. Wong, N. Y. Yang, “Dependency analysis of accuracy estimates in k-fold cross validation”, IEEE Transactions on Knowledge and Data Engineering, vol. 29, no. 11, 2017.
• 14. M. F Moller, “A scaled conjugate gradient algorithm for fast supervised learning”, Neural Networks, vol. 6, no. 4, pp. 525-533, 1993.
• 15. P. T. de Boer, D. P. Kroese, S. Mannor R. Y. Rubinstein, “A tutorial on the cross-entropy method”, Annals of Operations Research, vol. 134, no. 1, pp. 19-67, 2005.
• 16. T. Kautz, B. M. Eskofier, C. F. Pasluosta, “Generic performance measure for multiclass-classifiers”, Pattern Recognition, vol. 68, pp. 111-125, 2017.
• 17. M. Nabian, “A comparative study on machine learning classification models for activity recognition”, J Inform Tech Softw Eng, vol. 7, no. 4, p. 209, 2017.