Intelligent Technologies for Research and Engineering

Volume: 2

Diagnose Breast Cancer on Mammography Using Self Supervised Decision Tree Algorithm

Author(s): A. Pon Bharathi*, A. S. Sarika, V. Suresh Babu, M. Jayaprakash and Allan J. Wilson

Pp: 135-156 (22)

DOI: 10.2174/9789815165586124020012

* (Excluding Mailing and Handling)

Abstract

Women in India are diagnosed with breast cancer at a far higher rate than men. Breast cancer is the most common kind of cancer in women, accounting for more than half of all cancer cases. It is possible to minimize the mortality rate of breast cancer via early and precise detection. By utilizing mammography, early breast cancer detection, and assessment is currently possible. There is still a lot of controversy about mammography mass classifications, yet they are crucial for helping radiologists make accurate diagnoses. Using convolutional neural networks (CNNs), it has become possible to classify and segment images in a meaningful way. Unlabelled picture data, on the other hand, presents challenges, and while manual labelling is inefficient, pretrained CNNs also perform poorly on genuine medical images. In this research, we propose the use of Transformer-Based Networks (TBN) in computer vision. Transformer-based vision models have been found to outperform convolutional models in previous studies as well. A self-supervised learning (SSL) technique called the Decision Tree Algorithm (DTA) is proposed in this study for processing mammography images for diagnostic purposes. The Decision Tree Algorithm works effectively with categorical and continuous dependent variables. In this study, the population was divided into two or more homogeneous groups based on the most important traits and independent factors. According to this article, a previously trained model can be enhanced by transitioning from making predictions on uniformly tiled regions to making predictions on the complete image. There were two studies that utilized the Kaggle archive breast cancer sample pool, the second of which used 286 samples. In the initial experiment, the decision tree was 100 percent accurate, but in the follow-up investigation, it was only 97.9 percent accurate.

© 2024 Bentham Science Publishers | Privacy Policy