SRMADNet: Swin ResUnet3+-based mammogram image segmentation and heuristic adopted multi-scale attention based DenseNet for breast cancer detection

• Published in: Biomedical signal processing and control Vol. 88; p. 105515
• Main Authors: Ghuge, Kalyani, Saravanan, Dr. D.
• Format: Journal Article
• Language: English
• Published: Elsevier Ltd 01.02.2024
• Subjects: Adaptive multi-scale attention-based densenet with extreme learning machine; Breast cancer classification, mammogram image; Position-based improved invasive weed and crisscross optimization; Swin Resunet3; Adaptive multi-scale attention-based densenet with extreme learning machine; Swin Resunet3; Position-based improved invasive weed and crisscross optimization; Breast cancer classification, mammogram image
• ISSN: 1746-8094
• DOI: 10.1016/j.bspc.2023.105515

•To develop the architecture of Swin ResUnet3+-based Mammogram Image Segmentation for breast cancer classification model.•To segment the abnormal regions using Swin ResUnet3+, where it entails a Swing transformer, residual network, and Unet3+.•To create the Adaptive Multi-scale Attention-based Densenet with Extreme Learning Machine (AMAD-ELM) for classification.•To adopt the Position-based Improved invasive Weed and Crisscross Optimization (PIWCO) for tuning the factors in AMAD-ELM. Breast tumor is the second wide largest mortality rate of all the other cancers that are found in women and it is also thought to be the main factor in a high death rate. Mammographic pictures are mostly examined by the trained and expert radiologists to identify the abnormalities in the breast such as microcalcifications and masses. A correct diagnosis and early detection can aid in the fight against this form of dangerous disease. Breast cancer identification and categorization have been found to benefit greatly from the segmentation of mammography lesions. Segmentation can help with both shape-related feature extraction and precise lesion localization. Therefore, specialists implemented the latest image segmentation for mammogram images and the identification architecture for the identification of breast cancers, most importantly for the women. The wanted mammogram pictures are collected from traditional global databases. The gathered pictures are fed into the segmentation process, where the images are segmented employing Swin ResUnet3+. Thus, the mammogram segmented pictures are forwarded to the identification stage, where Adaptive Multi-scale Attention-based Densenet with Extreme Learning Machine (AMAD-ELM) model is implemented to identify the breast tumor effectively from the mammogram pictures. The parameters from AMAD-ELM are optimized through the suggested Position-based Improved invasive Weed and Crisscross Optimization (PIWCO). The effectiveness of the recommended breast tumor identification system with the help of deep learning is analyzed through various existing models to show its effectiveness.