Atrous Convolution-Based Adaptive 3D-CNN Model for Breast Cancer Diagnosis Using Segmentation in Mammogram Images

A commonly affected disease for women is breast cancer, caused by abnormal growth of the breast tissues. Existing breast cancer detection approaches rely on manual segmentation, which consumes more time and is ineffective for handling variations in breast tissue density and texture. Moreover, these methods often fail to detect subtle abnormalities, leading to missed diagnoses. Additionally, they are vulnerable to overfitting, lack robustness to noise and artifacts, and require extensive computational resources. To overcome these challenges, a novel deep learning-based framework for breast cancer detection using mammogram images has been introduced. At first, the necessary images are collected from online sources. The mammogram image is subjected to the preprocessing approach via contrast-limited adaptive histogram equalization (CLAHE) and histogram equalization (HE) to obtain a high-contrast image with reduced noise. Consequently, the pre-processed image is fed to the hybridization of Improved UNet-FCN for segmenting the image over cancer-occurred regions. The parameters within the segmentation are optimized by the farmland fertility snow leopard optimization (FFSLO). After, the segmented image is given to the breast cancer detection stage. Here, the atrous convolution-based adaptive 3D-convolutional neural network (AC-A3DCNN) is utilized to detect breast cancer. Here, the variable tuning is carried out with the FFSLO to boost the demonstrated approach’s detection accuracy rate. While validating the statistical test, the designed model shows 2.61%, 2.05%, 4.86%, and 0.68% elevated than CHOA, CMO, FFOA, and SNOA for best metrics. Hence, the offered approach’s efficacy is revealed through the comparative evaluation of the diverse baseline models concerning the standard performance measures.