Classification of healthy and unhealthy patients in dynamic infrared thermography of breasts using recurrent neural networks

• Antonio Borralho UFMA

Resumo

Breast cancer is the second most common type of cancer, being associated with a high mortality rate of women in the world. This paper proposes a methodology to classify patients as healthy or not healthy at images of infrared thermography dynamics, modeling the problem as time series. Techniques were used for image processing and analysis of time series through the use of recurrent neural networks such as long short-term memory. The data set is constituting 70 exams, 35 of them ill and 35 healthy patients. The proposed methodology has reached 92.3% of recall. Despite the result will not be close to the state-of-art, it is believed that this can be improved with the generation of synthetic series.

Referências

Borchartt, T. (2013). Análise de imagens termográficas para a classificação de alterações na mama. Federal Fluminense University.

Bozek, J. et al. (2009). A Survey of Image Processing Algorithms in Digital Mammography. Recent Advances in Multimedia Signal Processing and Communications.

Brown, L. G. (2004). A survey of image registration techniques. ACM Computing Surveys (CSUR).

Brownlee, J. (2019). Machine Learning Algorithms From Scratch With Python. Machine Learning Mastery Pty. Ltd., 1th edition.

Cheng, H. D. et al. (2006). Approaches for automated detection and classification of masses in mammograms. Pattern recognition.

Forestier, G. (2016). Generating synthetic time series to augment sparse datasets. Data Mining (ICDM).

Silva, L. F. (2015). Uma Análise Hı́brida para Detecção de Anomalias da Mama usando Séries Temporais de Temperatura. PhD thesis, Federal Fluminense University, Rio de Janeiro.

Sutskever, I. (2013). Training Recurrent Neural Networks. PhD thesis, University of Toronto, Canada.