THE APPLICATION OF AI-BASED TECHNIQUES FOR EARLY DETECTION OF BREAST CANCER
Dafina Xhako, Elda Spahiu, Suela Hoxhaj, Niko Hyka · RAP Conference Proceedings · 2024
Breast cancer is a type of tumor that occurs in breast tissue.It continues to remain one of the most prevalent and life-threatening diseases globally, becoming the second leading cause of cancer-related deaths among women.Breast cancer begins when malignant and cancerous cells begin to grow from the breast cells.Self-tests and periodic clinical examinations help in early diagnosis and significantly improve survival chances.Early diagnosis of breast cancer, when it is small and has not spread, can make the disease easier to treat, thus increasing the patient's chances of survival.Due to the medical importance of breast cancer examinations, Computer-Aided Detection methods have been developed to detect anomalies such as calcifications, masses, architectural distortions, and bilateral asymmetry.Micro calcifications are nothing but tiny mineral deposits within the breast tissue.They look like small white colored spots.They may or may not be caused by cancer.This is one reason why breast cancer detection is difficult with mammogram because the mammogram results vary greatly depending on the patient's age, breast density, and the type of lesion present.Breast density can lead to differences in the contrast of malignant regions and can lead to incorrect conclusions.Our study describes an AI approach of adaptive median filter which performs spatial processing to determine which pixels in an image have been affected by noise.To detect a tumor at different stages we use neural network with different learning techniques to get Gaussian Mixed Model (GMM) segmentation.The Artificial Neural Network (ANN) model is based on convolutional neural networks (CNN) and as input data we have selected 260 mammogram images classifying them into three categories: normal mammogram, mammogram with benign and mammogram with cancer.After the training process, we used a CNN model named ResNet50 to compare the results.Due to the low processing capacity, we have chosen a small dataset.Our results show that a CNN model with 3*3 convolutional layer performed better compared with Gaussian Mixed Model segmentation.