Deep Learning Algorithms for Diagnosis of Breast Cancer with Maximum Likelihood Estimation

Cifci, Mehmet Akif; Aslan, Zafer

Machine Learning (ML) and particularly Deep Learning (DL) continue to advance rapidly, attracting the attention of the health imaging community to apply these techniques to increase the precision of cancer screening. The most common cancer in women is breast cancer that affects more than 2 million women each year and causes the largest number of deaths from cancer in the female population. This work provides state-of-the-art research on the contributions and new applications of DL for early diagnosis of breast cancer. Also, it emphasizes on how and which major applications of DL algorithms are going to be benefitted for early diagnosis of breast cancer for which CNNs, one of the DL architectures, will be used. In this study, a DL method to be used for diagnostic and prognostic analysis using the X-ray breast image dataset for breast cancer is studied. Based on the dataset, it is aimed to diagnose breast cancer at an early stage. Thus, it may take place before a clinical diagnosis. For the testing probability of the disease, 21400 X-ray breast images, both normal and cancer, were taken from USF mammography datasets. From these images, 70% is used as the training step, while 30% of images are benefitted for the testing step. After the implementation of the architecture, VGG16 has achieved an overall accuracy of 96.77%, with 97.04% sensitivity and 96.74% as AUC, while Inception-v4 has an overall accuracy of 96.67%, with 96.03% sensitivity and 99.88% as AUC. These results show the high value of using DL for early diagnosis of breast cancer. The results are promising.