PWidFHNet: Parallel Wider Forward Harmonic Net to Classify Breast Cancer Using Histopathological Images.

BACKGROUND/PURPOSE: Breast cancer poses a significant health risk to women worldwide. The timely diagnosis of breast cancer is essential to increase the survival rate and to control premature mortality. However, existing classification techniques may not account for all the variations of breast cancer, leading to potential misclassification, which adversely affects treatment decisions. This article, therefore, puts forward the PWidFHNet model to enable accurate classification of breast cancer from histopathological images. METHODS: Initially, the acquired histopathological image is subjected to image preprocessing using a Gaussian filter. Subsequently, blood cell segmentation is performed using the Tversky-parallel reverse attention network. Moreover, feature extraction is performed, where shape features and learned invariant feature transform are extracted. Finally, the proposed PWidFHNet is used to perform the breast cancer classification. Here, PWidFHNet is established by incorporating the parallel convolutional neural network with the wide residual network, in which the network layers are enhanced through harmonic analysis. RESULTS: The performance of PWidFHNet is evaluated using metrics such as specificity, accuracy, and sensitivity, all of which achieved their maximum values of 92.76%, 91.60%, and 90.86%, respectively. CONCLUSION: The proposed PWidFHNet model provides a reliable and efficient framework for breast cancer classification by integrating advanced image pre-processing, segmentation, feature extraction, and deep learning techniques. The achieved performance highlights its potential as a diagnostic support tool to assist pathologists in breast cancer assessment and clinical decision-making.