| dc.description.abstract | Breast cancer is the most prevalent disease and leading cause of death among women.
Mammography plays an important role in early detection, and effective breast cancer
screening contributes to disease prevention. Due to a lack of medical resources, manual
diagnosis is time-consuming, error-prone, and complex in clinical settings, and can lead to
deterioration of patient condition and delay in treatment. Automated systems that detect
abnormalities on medical images in order to assist physicians have become a hot research
topic in the medical industry. The rapid development of deep learning has piqued the
interest of researchers in medical imaging. I proposed to develop a deep learning model for
key detection using a training strategy that effectively uses the extension of an annotated
training dataset or simply the cancer status of the entire image. Identify benign and
malignant abnormalities on mammograms used in automated diagnostics. InceptionV3,
ResNet50, VGG16, and MobileNetV2 models are proposed to be tested on three public
datasets, DDSM, CBIS-DDSM, and MIAS, in order to classify mammography
abnormalities. When the InceptionV3 architecture was refined, the DDSM dataset was
analyzed with the highest accuracy (92.78%). The maximum area under the ROC curve
(AUC) is 92,33 percent. This is the result of combining additional data with breast
segmentation. The purpose of this study was to develop a deep learning model capable of
identifying breast cancer by differentiating benign and malignant tasks on mammograms
of varying densities, and to compare the models' performance to that of previous research. | en_US |
