Robustez de las redes neuronales profundas para clasificar imágenes médicas. / Robustness of deep neural network to classify medical images.

Kloster, Matias A. (2019) Robustez de las redes neuronales profundas para clasificar imágenes médicas. / Robustness of deep neural network to classify medical images. Integration Project in Mechanical Engineering, Universidad Nacional de Cuyo, Instituto Balseiro.

PDF (Tesis)

Abstract in Spanish

En esta tesis se aborda el problema de clasificación automática de imágenes utilizando técnicas de redes neuronales profundas. Estas técnicas permiten lograr resultados comparables con los del sistema visual humano, pero presentan limitaciones tales como ser vulnerables a ejemplos adecuadamente elegidos. Estos ejemplos, son llamados ejemplos adversariales y se pueden construir de diversas maneras a partir de una imagen natural. Los ejemplos adversariales son clasificados erróneamente por la red, pero son visualmente indistinguibles de las imágenes naturales. Luego de un estudio exhaustivo de dicha problemática, se realiza la generación de ejemplos adversariales sobre dos bases de datos ampliamente conocidas. Luego, dada una imagen se implementa un algoritmo que detecta si la misma es natural o adversarial, tomando nota de la precisión del método. Por ultimo, se realizaron las mismas pruebas sobre una base de datos de imágenes de una patología denominada retinopatía diabetica, en donde la red neuronal utilizada categoriza la imagen según la gravedad de la enfermedad o la ausencia de la misma. La detección de las imágenes adversariales de retinopatía diabetica utilizando el método propuesto en este estudio presenta un error de 3:8 %.

Abstract in English

This thesis approaches the problem of automatic image classification using deep neural network techniques. These techniques achieve results comparable with those of the human visual system, but they present limitations such as being vulnerable to properly chosen examples. These examples, are called adversarial examples and can be constructed in various ways from a natural image. Adversarial examples are erroneously classified by the network, but are visually indistinguishable from natural images. After an exhaustive study of this problem, adversarial examples are generated on the basis of two widely known databases. Then, given an image, an algorithm is implemented that detects if it is natural or adversarial, taking note of the precision of the method. Finally, the same tests were performed on a database of images of a pathology called diabetic retinopathy, where the neuronal network categorizes the image according to the severity of the disease or its absence. The detection of diabetic retinopathy adversarial images using the method proposed in this study presents an error of 3:8%.

Item Type:Thesis (Integration Project in Mechanical Engineering)
Keywords:Neural networks; Redes neuronales; Pathology; Patología; [Robustness; Robustez; Deep machine learning: Aprendizaje automático]
References:[1] Russell, S. J., Norvig, P. Inteligencia Articial: un enfoque moderno. 2 edición. Prentice Hall, 2008. 1, 3 [2] Sermanet, P., Eigen, D., Zhang, X., Mathieu, M., Fergus, R., LeCun, Y. Overfeat: Integrated recognition, localization and detection sing convolutional networks. arXiv preprint arXiv:1312.6229, 2013. 2 [3] Wu, Y., Giger, M. L., Doi, K., Vyborny, C. J., Schmidt, R. A., Metz, C. E. Articial neural networks in mammography: application to decision making in the diagnosis of breast cancer. Radiology, 187 (1), 81-87, 1993. 2 [4] Esteva, A., Kuprel, B., Novoa, R. A., Ko, J., Swetter, S. M., Blau, H. M., et al. Dermatologist-level classication of skin cancer with deep neural networks. Nature, 542 (7639), 115, 2017. 2 [5] Pratt, H., Coenen, F., Broadbent, D. M., Harding, S. P., Zheng, Y. Convolutional neural networks for diabetic retinopathy. Procedia Computer Science, 90, 200-205, 2016. 2 [6] Szegedy, C., Zaremba, W., Sutskever, I., Bruna, J., Erhan, D., Goodfellow, I., et al. Intriguing properties of neural networks. arXiv preprint arXiv:1312.6199, 2013. 2, 3, 13 [7] Goodfellow, I. J., Shlens, J., Szegedy, C. Explaining and harnessing adversarial examples. arXiv preprint arXiv:1412.6572, 2014. 2, 3, 13 [8] Diabetic retinopathy detection - kaggle. URL Accedido 25-05- 2019. 3, 28, 29 [9] James, G., Witten, D., Hastie, T., Tibshirani, R. An introduction to statistical learning, tomo 112. Springer, 2013. 3 [10] Hurwitz, J., Kirsch, D. Machine learning for dummies - IBM limited edition. John Wiley & Sons, Inc, 2018. 3, 5, 6 [11] Hertz, J. A. Introduction to the theory of neural computation. CRC Press, 2018. 3, 6, 7, 8, 9, 11 [12] Moosavi-Dezfooli, S.-M., Fawzi, A., Frossard, P. Deepfool: a simple and accurate method to fool deep neural networks. En: Proceedings of the IEEE conference on computer vision and pattern recognition, págs. 2574-2582. 2016. 3, 12, 13, 15, 16, 34 [13] Carlini, N., Wagner, D. Adversarial examples are not easily detected: Bypassing ten detection methods. En: Proceedings of the 10th ACM Workshop on Articial Intelligence and Security, pags. 3{14. ACM, 2017. 3, 17 [14] Conda documentation. URL Accedido 23-05-2019. 3 [15] NVIDIA. Nvida cudnn | nvidia developer. URL Accedido 23-05-2019. 4 [16] Keras. URL Accedido 23-05-2019. 4, 7, 29 [17] Google. Tensor ow. URL https://www.tensor Accedido 23-05-2019. 4 [18] He, K., Zhang, X., Ren, S., Sun, J. Delving deep into rectiers: Surpassing humanlevel performance on imagenet classication. En: Proceedings of the IEEE international conference on computer vision, pags. 1026{1034. 2015. 6 [19] Rosenblatt, F. The perceptron: a probabilistic model for information storage and organization in the brain. Psychological review, 1958. 8 [20] Minsky, M., Papert, S. Perceptron: an introduction to computational geometry. The MIT Press, Cambridge, expanded edition, 19 (88), 2, 1969. 8 [21] Cybenko, G. Approximation by superpositions of a sigmoidal function. Mathe-matics of control, signals and systems, 2 (4), 303{314, 1989. 8 [22] Bryson, A. E., Ho, Y. C. Applied Optimal Control. New York: Blaisdell, 1969. 9 [23] Werbos, P., J. (Paul John, P. Beyond regression : new tools for prediction and analysis in the behavioral sciences. Ph.D. Thesis, Harvard University, 1974. 9 [24] Parker, D. Technical report tr-47 , center for computational research in economics and management science, massachusetts institute of technology, cambridge. MA, 1985. 9 [25] Rumelhart, D. E., Hinton, G. E., Williams, R. J. Learning internal representations by error propagation. Inf. téc., California Univ San Diego La Jolla Inst for Cognitive Science, 1985. 9 [26] LeCun, Y. Une procedure d'apprentissage ponr reseau a seuil asymetrique. Pro-ceedings of Cognitiva 85, pags. 599-604, 1985. 9 [27] Fukushima, K. Neocognitron: A self-organizing neural network model for a mechanism of pattern recognition unaected by shift in position. Biological cybernetics, 36 (4), 193-202, 1980. 9 [28] Hubel, D. H., Wiesel, T. N. Receptive elds of single neurones in the cat's striate cortex. The Journal of physiology, 148 (3), 574-591, 1959. 9 [29] Hubel, D. H., Wiesel, T. N. Receptive elds and functional architecture of monkey striate cortex. The Journal of physiology, 195 (1), 215{243, 1968. 9 [30] LeCun, Y., Boser, B., Denker, J. S., Henderson, D., Howard, R. E., Hubbard, W., et al. Backpropagation applied to handwritten zip code recognition. Neural computation, 1 (4), 541-551, 1989. 9 [31] Goodfellow, I., Bengio, Y., Courville, A. Deep Learning. MIT Press, 2016. http // 10 [32] LeCun, Y., Bottou, L., Bengio, Y., Haner, P., et al. Gradient-based learning applied to document recognition. Proceedings of the IEEE, 86 (11), 2278{2324, 1998. 10, 11 [33] The mostly complete chart of neural networks. URL Accedido 05-06-2019. 11 [34] Su, J., Vargas, D. V., Sakurai, K. One pixel attack for fooling deep neural networks. IEEE Transactions on Evolutionary Computation, 2019. 13 [35] Feinman, R., Curtin, R. R., Shintre, S., Gardner, A. B. Detecting adversarial samples from artifacts. arXiv preprint rXiv:1703.00410, 2017. 17 [36] Kappa. URL Accedido 27-05-2019. 29
Subjects:Physics > Redes neuronales
Medicine > Física médica
Divisions:Gcia. de área de Investigación y aplicaciones no nucleares > Gcia. de Física > Física médica
ID Code:836
Deposited By:Tamara Cárcamo
Deposited On:15 Mar 2021 11:52
Last Modified:15 Mar 2021 11:52

Repository Staff Only: item control page