부산대학교 도서관

Breast cancer is widespread throughout the world and can be cured if diagnosed early. Mammography is an irreplaceable and critical technique in modern medicine that serves as a foundation for the detection of breast cancer. In medical imaging, the reliability of synthetic mammogram images is produced by deep convolutional generative adversarial networks (DCGAN). Human validation to assess the quality of synthetic images to examine and calculate the perceptual variations between synthetic images and their real-world counterparts is a difficult task. Thus, this research focused on improving the quality and authenticity of synthetic mammogram images. For this, we explored and identified a new research gap because radiologists consistently expressed much higher confidence levels in real mammogram images in their assessment process. This research highlights the key difference between synthetic and real mammograms by defining mean scores. The defined mean identifies a large gap, with real mammographic images receiving an average score of 0.73 and a synthetic score of 0.31. A statistical analysis was performed, which produced a T-statistic of -6.35, a p-value less than 0.001, and a 95% confidence interval ranging from -0.50 to -0.28. These results have a wide range of implications. It emphasizes the urgent need for further improvements in the generative model, improving the legitimacy and caliber of synthetic mammogram images. Our research highlights how crucial it is to incorporate synthetic images into clinical practice with caution and thought. Ethical considerations must encompass the potential consequences of relying on synthetic data in medical decision-making, along with concerns related to diagnostic accuracy and patient safety.