Classification of breast cancer with deep learning from noisy images using wavelet transform

Area of Science:

• Medical Imaging
• Computational Pathology
• Artificial Intelligence in Medicine

Background:

• Histopathological images are crucial for breast cancer diagnosis.
• Image noise can significantly degrade diagnostic accuracy.
• Deep learning models require high-quality images for reliable classification.

Purpose of the Study:

• To evaluate the effectiveness of Wavelet Transform (WT) for denoising histopathological images.
• To compare the performance of a proposed Convolution Neural Network (CNN) model against VggNet-16 for breast cancer classification.
• To identify optimal noise parameters for image preprocessing in breast cancer detection.

Main Methods:

• Histopathological images were artificially corrupted with various noise types and intensities.
• Wavelet Transform (WT) was applied to denoise the corrupted images.
• Denoised images were classified using a proposed CNN model and the VggNet-16 architecture.
• Peak Signal to Noise Ratio (PSNR) was used to evaluate denoising performance.

Main Results:

• Wavelet Transform effectively reduced noise in histopathological images.
• Gaussian noise removal yielded superior Peak Signal to Noise Ratio (PSNR) values compared to other noise types.
• The proposed CNN model achieved a classification accuracy of 86.9%, outperforming the VggNet-16 model.
• Optimal performance was achieved with Gaussian noise at 0.3 intensity.

Conclusions:

• Wavelet Transform is an effective preprocessing step for enhancing histopathological image quality.
• The proposed CNN model demonstrates superior performance for breast cancer classification compared to VggNet-16.
• Optimized noise reduction techniques can significantly improve the accuracy of AI-driven cancer diagnostics.