Breast Cancer Detection Using Thermal Images and Clinical Data: Anatomy Hover Multi-Scale Random-Shape Convolution Neural Network With Salp Swarm Optimization

Worldwide, breast cancer (BC) ranks as the second leading cause of mortality for women; increasing survival rates require an early and precise diagnosis. Despite the promising results of deep learning (DL) algorithms in BC detection, the majority of existing methods have a high false-positive rate, low precision, and misdiagnosis. These complications negatively affect the quality of clinical decision-making and underline the necessity to find more accurate diagnostic solutions. A novel Anatomy Hover Multi-Scale Random-Shape Convolution Neural Network with Salp Swarm Optimization (AHMSRSCNN-SSO) has been proposed to overcome these difficulties by detecting BC using clinical data and thermal images from the DMR and Breast thermography dataset from Kaggle. Gradient Domain Weighted Guided Image Filtering (GDWGIF) is used as a pre-processing approach to improve the quality of the images. Inverse Z-transform and Wiener-Hopf factorization are then used for segmentation. The classification is performed using AHMSRSCNN, which integrates Anatomy-Aware Hover-Transformer (AHT) and Multi-Scale Random-Shape Convolutional Neural Network (MSRSCNN) components for improved accuracy. Salp Swarm Optimization (SSO) is utilized for adaptive weight adjustment during training to optimize learning. Experimental findings indicate a very high accuracy (99.9%) and precision (99.8%), which proves that the proposed methodology is better than the available methods and it is a promising development in the area of thermal image-based BC detection.