Performance Analysis of Deep Learning Models over BreakHis Dataset using Up-Sampling and Down-Sampling Techniques for Classification of Breast Cancer

• Published in: 2023 9th International Conference on Smart Computing and Communications (ICSCC) pp. 594 - 599
• Main Authors: Ahirwar, Madhvendra, Agrawal, Anupam
• Format: Conference Proceeding
• Language: English
• Published: IEEE 17.08.2023
• Subjects: Breast cancer; Computational modeling; Convolutional Neural Network (CNN); Data augmentation; Deep learning; Down-Sampling Breast Cancer Classification; Geometry; Image processing; Up-Sampling; Visual Geometry Group 16 (VGG-16); Visualization
• DOI: 10.1109/ICSCC59169.2023.10334935

It's very challenging to determine the actual cause of breast cancer, therefore early diagnosis is crucial to lowering the disease's fatality rate. The likelihood of survival increases by up to 8% with early cancer identification. Even seasoned radiologists, however, struggle to recognize characteristics like micro-classifications, lumps, and masses, which results in significant false positive and false negative rates. In recent years we have seen a lot of development in deep learning and image processing which gives rise to some optimism for the creation of improved applications for the early diagnosis of breast cancer. In my work, the performance of a Convolutional Neural Network (CNN) and Visual Geometry Group 16 (VGG-16) was analyzed on a breast cancer dataset using upsampling and downsampling techniques. Upsampling involves increasing the number of instances in the minority class (in this case, breast cancer cases) by duplicating them or by performing some other data augmentation technique, while downsampling involves reducing the number of instances in the majority class (non-breast cancer cases) by randomly removing some of them. The purpose of these techniques is to address imbalanced datasets, where one class significantly outnumbers the other. The results of the analysis showed that CNN performed better on the balanced datasets achieved through upsampling and downsampling.