A Novel Transfer Learning Technique for Detecting Breast Cancer Mammograms Using VGG16 Bottleneck Feature

• Published in: ECS transactions Vol. 107; no. 1; pp. 733 - 746
• Main Authors: Prusty, Sashikanta, Dash, Sujit Kumar, Patnaik, Srikanta
• Format: Journal Article
• Language: English
• Published: The Electrochemical Society, Inc 24.04.2022
• ISSN: 1938-5862; 1938-6737
• DOI: 10.1149/10701.0733ecst

Breast cancer represents the highest percentage of cancers and the second most common cancer overall that affect women with 87,090 deaths approximately as reported by ICMR, 2018 in India (1). Breast tumours are classified into two types as a) benign, which is not very harmful and would not cause breast cancer and b) malignant, the tumours are extremely dangerous and would form an abnormal cell that may cause cancer. One transfer learning model called VGG16 (Visual Geometry Group (16) has been implemented with a focus on breast cancer classification using mammography images taken from the MIAS dataset. Moreover, at the preprocessing step, all the breast images are enhanced by using CLAHE (Contrast Limited Adaptive Histogram Equalization) technique that specifically maintains adaptive techniques to remove the lines, letters, and other boxes irrelevant to the Breast Image. As the name suggests, this VGG16 has taken 13 convolutional layers and 3 fully connected layers to train the dataset. In the training period, this model could predict whether the breast images contain any type of cancerous cells or not. And finally, the model has successfully implemented and produced Network’s test score of 87.999 percent.