A Deep Learning-based Model for Predicting Survivability of Breast Cancer Patients. 

“Cancer disease is one of the major health problems worldwide. Breast cancer is one of the most common malignancies in women. Early diagnosis of Breast cancer patients based on an accurate prediction system can increase the survival of the patients to rate of 86%. The prediction of survivability is one of the most challenging for healthcare professionals. Therefore, studying the predication survivability of breast cancer patients is very important to help healthcare professionals to make informed decisions on the potential necessity of adjuvant treatment. In addition, giving the patients, who most likely will survive Cancer, hope will improve their psychological health.

Machine learning (ML) is widely used in this area, several algorithms that have been used frequently in the previous studies and have given varying results to their success ratio such as Decision Tree, Support vector machine, and Naïve Bayes algorithm. One of the most promising subsets of ML is deep learning algorithms that achieve remarkable performance in different areas such as bioinformatics, speech recognition, Computer vision, and Deep-learning robots, etc.

In this thesis, two Deep Learning Algorithms were employed to investigate their capability in predicting the survivability of breast cancer patients using Deep Neural Network (DNN) and Convolutional Neural Network (CNN).

This thesis addresses the factors that might have an impact on the performance of the two algorithms and then compares between them and between other ML algorithms to determine which may give better results in terms of their capabilities in predicting the survivability of breast cancer patients.

To evaluate the studied algorithms, the breast cancer dataset is used from the Surveillance, Epidemiology, and End Results (SEER) program. The experiment results show that the DNN algorithm achieves prediction superiority over the CNN algorithm and other ML algorithms. To evaluate the performance of the prediction models, accuracy, sensitivity, and specificity metrics were used. The accuracy value obtained by the DNN algorithm is 86.44%, while the sensitivity value obtained is 76.18%, and the specificity value is 94.48%. Changing the architecture of the algorithms has a positive impact on improving the results, the three main factors are the number of layers, numbers of the epoch, the batch size; these factors were tuned until we reach to the optimal number for each of them. Finally, these results indicate that the DNN prediction model could be significantly used for survivability prediction suggesting time- and cost-effective treatment for breast cancer patients and could be translated into decision support tools in the medical domain.” 

Keywords: Survivability of Breast Cancer, Machine Learning, Deep Learning, Deep Neural Network, Convolutional Neural Network. Surveillance, Epidemiology, and End Results (SEER)

​​