Study of Models based on Deep Neural Networks for Conjunctival Melanocytic Tumor Classification

• Rafael B. dos Santos UEFS
• Matheus G. Pires UEFS
• Fabiana C. Bertoni UEFS

DOI: https://doi.org/10.5753/sbcas.2024.2762

Abstract

Conjunctival melanoma is a malignant neoplasm which generally presents as a pigmented nodular conjunctival lesion. Variant cases with several atypical shapes can delay the diagnosis. To assist the doctor in early diagnosis, minimizing risks to the patient, this work conducted a comparative study of algorithms to classify conjunctival melanoma. For this purpose, models based on Convolutional Neural Networks were evaluated in binary and multiclass classification of tumors, based on VGG16, Xception and MobileNetV2 models, using the Transfer Learning technique to improve generalization. For final image classification, an approach based on an assembly of classifiers was performed, consisting of the PMC, SVM and KNN algorithms. The study used a dataset with 406 images, applying data balancing techniques, such as SMOTE and ADASYN. To find the best classification model, was used 5-folds cross validation technique. Considering all the tests carried out, Ensemble MobileNetV2 models was the one that obtained the best results.

References

Allam, E., Salem, A., and Alfonse, M. (2024). Classification of orbital tumors using convolutional neural networks. Neural Computing and Applications, 36:6025–6035.

Chandrabhatla, A., Horgan, T., Cotton, C., Ambati, N., and Shildkrot, Y. (2023). Clinical applications of machine learning in the management of intraocular cancers: A narrative review. Investigative ophthalmology visual science, 64:29.

Chawla, N. V., Bowyer, K. W., Hall, L. O., and Kegelmeyer, W. P. (2002). Smote: Synthetic minority over-sampling technique. The Journal of Artificial Intelligence Research, 16:321–357.

Chollet, F. (2017). Xception: Deep learning with depthwise separable convolutions.

Fawcett, T. (2006). An introduction to roc analysis. Pattern Recognition Letters, 27(8):861–874. ROC Analysis in Pattern Recognition.

He, H., Bai, Y., Garcia, E. A., and Li, S. (2008). Adasyn: Adaptive synthetic sampling approach for imbalanced learning. In 2008 IEEE International Joint Conference on Neural Networks (IEEE World Congress on Computational Intelligence), pages 1322–1328.

Jain, P., Finger, P. T., Fili, M., Damato, B., Coupland, S. E., Heimann, H., Kenawy, N., Brouwer, N. J., Marinkovic, M., Duinen, S. G. V., Caujolle, J. P., Maschi, C., Seregard, S., Pelayes, D., Folgar, M., Yousef, Y. A., Krema, H., and Calle-Vasquez, B. G. A. (2021). Conjunctival melanoma treatment outcomes in 288 patients: a multicentre international data-sharing study. British Journal of Ophthalmology, 105(10):1358–1364.

Koseoglu, N., Corrêa, Z., and Liu, T. Y. (2023). Artificial intelligence for ocular oncology. Current opinion in ophthalmology, Publish Ahead of Print.

Li, Z., Qiang, W., Chen, H., Pei, M., Yu, X., Wang, L., Li, Z., Xie, W., Wu, X., Jiang, J., and Wu, G. (2022). Artificial intelligence to detect malignant eyelid tumors from photographic images. npj Digital Medicine, 5:23.

Nazir, S., Dickson, D. M., and Akram, M. U. (2023). Survey of explainable artificial intelligence techniques for biomedical imaging with deep neural networks. Computers in Biology and Medicine, 156:106668.

Novais, G. A. and Karp, C. L. (2012). Redes neurais profundas e ensemble de classificadores: uma aplicação em imagens médicas. Arquivos Brasileiros de Oftalmologia, 75:289–295.

Sandler, M., Howard, A., Zhu, M., Zhmoginov, A., and Chen, L.-C. (2018). Mobilenetv2: Inverted residuals and linear bottlenecks. In Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition (CVPR).

Santos-Bustos, D. F., Nguyen, B. M., and Espitia, H. E. (2022). Towards automated eye cancer classification via vgg and resnet networks using transfer learning. Engineering Science and Technology, an International Journal, 35:101214–101226.

Simonyan, K. and ZissermanK, A. (2014). Very deep convolutional networks for largescale image recognition. CoRR, pages 1–14.

Srivastava, N., Hinton, G., Krizhevsky, A., Sutskever, I., and Salakhutdinov, R. (2014). Dropout: a simple way to prevent neural networks from overfitting. J. Mach. Learn. Res., 15(1):1929–1958.

Yoo, T. K., Choi, J. Y., Kim, H. K., Ryu, I. H., and Kim, J. K. (2021). Adopting low-shot deep learning for the detection of conjunctival melanoma using ocular surface images. Computer Methods and Programs in Biomedicine, 205:1–10.