May 2024
International Journal of Information Technology
Binary Imaging Reporting and Data System (BIRADS) categorizes the mammogram masses using their shape, size, and density for the earlier detection of breast malignancy to reduce the mortality rate. For the efficient earlier detection, the proposed work develops a novel CNN with customized filters including High Frequency Boost Filter (HFBF) for extracting the complex features including sharp edges, sharp shapes and sharp patterns. The CNN with the customized filters is applied on the Discrete Wavelet Transform (DWT) sub-bands of Mammographic Image Analysis Society (MIAS) and Digital Database for Screening Mammography (DDSM) databases images with the fine-tuned hyperparameters. As a result, High-High (HH) sub-band images provide the highest accuracy of 97.28% and 97.94% on MIAS and DDSM databases respectively with stable measures of precision 0.96, recall 0.97 and F1-Score 0.97 on MIAS and precision 0.97, recall 0.98 and F1-Score 0.98 on DDSM when compared to the other sub-bands and original images of the databases. Likewise, the results obtained by the proposed model compared with other existing pre-defined models. The novel contribution of the proposed work is CNN architecture with customized filters for the efficient and complex feature extraction on the sub-bands of DWT. In addition, HH sub-band is considered for high frequency components of microcalcification rather neglating as noise (unlike other works).