Paper Title :Comparative Analysis of Deep Learning Techniques for Breast Tumor Classification on Small-Balanced and Large-Unbalanced Datasets

Author :Najme Zehra Naqvi, Amisha Jangra, Vittesha Gupta, Sanya Tiwari, Mahima Narang

Article Citation :Najme Zehra Naqvi ,Amisha Jangra ,Vittesha Gupta ,Sanya Tiwari ,Mahima Narang , (2023 ) " Comparative Analysis of Deep Learning Techniques for Breast Tumor Classification on Small-Balanced and Large-Unbalanced Datasets " , International Journal of Advance Computational Engineering and Networking (IJACEN) , pp. 73-79, Volume-11,Issue-6

Abstract : Effective screening of diseases is necessary in order to control them and prevent them from spreading. As computer-aided diagnosis technology continues to evolve, there is a discernible trend towards improving the accuracy of diagnoses. There are multiple requirements that biomedical image datasets must meet in order to be useful for implementing deep learning techniques for effective screening. However, biomedical image datasets are often small and limited. In this work, multiple research papers are reviewed and it is concluded that there is a lack of understanding regarding the effectiveness of deep learning techniques for such datasets. As a result, the accuracy of the implemented model suffers and biomedical images are not classified correctly. While there has been some research on this topic, it is still not clear whether deep learning techniques perform better or worse on small-balanced versus large-unbalanced datasets. To resolve this problem, we developed two different datasets from the Breast Ultrasound Images (BUSI) database - small and balanced containing equal number of images for all the classes and large but unbalanced using data augmentation. Then, a variety of deep learning techniques are used with for binary classification of ultrasound images of breasts tumours into malignant and benign classes. A comparative analysis of the implemented techniques is performed on both the datasets, small-balanced and the augmented dataset (large-unbalanced), by noting metrics accuracy score for each technique. Experimental findings have indicated that a greater accuracy is obtained for the small-balanced dataset using Mobile Net technique with an accuracy score of 87.5% as compared to 75% obtained using Dense Net method on the augmented-unbalanced dataset. Keywords - Breast Tumor Classification, Deep Learning, Image Classification

Type : Research paper

Published : Volume-11,Issue-6

	|		PDF	|	Viewed - 29	|	Published on 2023-10-13