Noise-Adaptive selective filtering for improved mammographic breast cancer detection using machine learning
|
Full Text |
Pdf
|
|
Author |
B. Maha Lakshmi, Pilla Bhanu Kiran, Kedarisetty Ayyan Balajee, Maharthi Aravind, Palisetti Anusha and Guru Akshaya
|
|
e-ISSN |
1819-6608 |
|
On Pages
|
382-389
|
|
Volume No. |
21
|
|
Issue No. |
6
|
|
Issue Date |
May 20, 2026
|
|
DOI |
https://doi.org/10.59018/032647
|
|
Keywords |
breast cancer, mammography, image noise, denoising, Gaussian noise, impulse noise, speckle noise, pre-processing, machine learning.
|
Abstract
Mammography is vital in screening early breast cancer, but the acquisition and transmission noise have a serious impact on reducing image quality and negatively in computer aided diagnosis (CAD) performance. In the real world, heterogeneous forms of noise are likely to be present in mammograms (including Gaussian, impulse, Poisson, and speckle noise), and fixed methods of denoising are thus inefficient. This paper suggests a noise-adaptive preprocessing system that first estimates the primary noise properties with the help of statistical measures, entropy, and a simple Random Forest categorizer, followed by selective filtering depending on the characteristics of the noise identified. Experiments conducted on the MIAS and CBIS-DDSM datasets demonstrate consistent improvements in both image quality and diagnostic classification. The proposed method achieves an average PSNR improvement of +10.2 dB, an SSIM increase of +0.26, and a classification accuracy of up to 92.3%, outperforming conventional single-filter preprocessing and competitive deep-learning denoisers. Statistical hypothesis testing confirms that the improvements are significant (p < 0.01). These results indicate that noise-aware adaptive preprocessing can substantially enhance mammographic CAD reliability and provide a practical foundation for future real-time clinical deployment.
Back