Abstract:

Breast cancer stands as a widespread ailment among women. Timely diagnosis and appropriate intervention significantly enhance prospects of recovery and thwart disease progression. Visual representations of breast cancer serve to delineate breast tissue into malignant and non-malignant regions, offering crucial insights, including the discernment of malignant cells, assessment of tumor dimensions, and staging of the neoplasm. Breast cancer screenings are implemented to ameliorate sensitivities, especially within the cohort of young and pregnant women. However, instances exist wherein the examination of radiographic portrayals may inadvertently overlook certain intricacies. Advanced artificial intelligence paradigms, coupled with sophisticated computational methodologies, are harnessed to attain outcomes of heightened precision. In this context, the segmentation methodology facilitated by ultrasound technology emerges as a pivotal intervention. In the present study, U-Net and an innovative Fully Convolutional Network architecture are utilized to identify deviations and demarcate carcinogenic zones. Within the scope of this research, the proposed Fully Convolutional Network architecture achieves a MeanIoU of 77.2%, precision rates of 71.6%, recall rates of 77.7%, and an F1 score of 74.5% in testing images. Conversely, the U-Net model yields comparable performance metrics, manifesting a MeanIoU of 76.4%, precision of 67.6%, recall of 80.4%, and an F1 score of 73%. Notably, the findings underscore the superiority of the proposed Fully Convolutional Network paradigm over the U-Net architecture. These results accentuate the significance of segmentation methodologies in the realm of breast cancer diagnosis and treatment, thus affirming the superior efficacy of the suggested Fully Convolutional Network architecture relative to the U-Net architecture.