A Comprehensive Analysis of Preprocessing Techniques for Thermal Breast Image Processing

Comprehensive and effective breast cancer screening programs are essential diagnostic instruments for early detection, which are then followed by rigorous intervention initiatives. A promising method for conducting non-invasive testing is the combination of remote sensing and thermal imaging technologies. Convolutional neural networks (CNNs) are capable of effectively identifying aberrant histological characteristics shared by most breast cancers; however, their application in breast cancer diagnosis is surprisingly limited. An overview of preprocessing techniques for thermal breast image processing is given in this paper. Several preprocessing techniques, including median filtering, wavelet transform, Wiener filtering, and histogram equalization, have been independently investigated in earlier research. There are very few all-inclusive techniques that methodically combine several conventional and statistical techniques to combine contrast enhancement and noise reduction in mammography images in the best possible way. Furthermore, there hasn't been much research done on using CNNs as preprocessing filters as opposed to classifiers. By developing a multi-step preprocessing pipeline that combines conventional filtering methods (Median, Wiener), DWT-based transformation techniques, and enhancement techniques (histogram equalization and dynamic edge sharpening), this study closes this knowledge gap. This study uses a detailed signal-to-noise ratio (SNR) analysis across frequency orientations to evaluate their combined impact on image quality.