Breast Cancer Diagnosis and Prognosis Using Artificial Neural Network

Abstract

As the mortality rate of breast cancer increases, there is the need to foster efforts to combat this ugly disease and save our mothers and all female folks from untimely death. Various methods being used to diagnosis this disease range from traditional hand sampling of the breast to a more advanced and generally accepted technique known as biopsy, a fine needle aspirations that enable early detection of the disease. Manual analysis of the biopsy image by clinical pathologist have many limitations, hence the need of computer aided approach. Extensive works had been carried out using this approach but more works still have to be done as a better model to rely on have not been achieved. In this thesis, we employed various deep neural network and machine learning frameworks like keras, tensorflow and lightgbm. We also made use of breast cancer histopathology image dataset release by Rui Yan et al, (2019) and Araujo et al, (2017). We deduced a model whose sensitivity score is higher than that of our redecessor. Reversed CRIP-DM methodology by John Rollins and very deep learning feature representation was used in the analysis. Lightgbm was used for the training and validation. At the end a python script was developed to perform real time breast cancer diagnosis.