Machine learning and Region Growing for Breast Cancer Segmentation
Diyar Qader Zeebaree, Habibollah Haron, Adnan Mohsin Abdulazeez, Dilovan Asaad Zebari · 2019
One of the main causes of increased mortality among women is breast cancer. The ultrasound scan is the most widely used method for diagnosing geological disease i.e. breast cancer. The first step for identifying the abnormality of the breast cancer (malignant from benign), is the extraction of the region of interest (ROI). In order to achieve this, a new approach to breast ROI extraction is proposed for the purpose of reducing false positive cases (FP). The proposed model was built based on the local pixel information and neural network. It includes two stages namely, training and testing. In the training stage, a trained model was built by extracting the number of batches from both ROI and background. The testing stage involved scanning the image with a fixed size window to detect the ROI from the background. Afterwards, a distance transform was used to identify the ROI and remove non-ROI. Experiments were conducted on the on-data set with 250 ultrasound images (150 benign and 100 malignant) the preliminary results show that the proposed method achieves a success rate of about 95.4% for breast contour extraction. The performance of the proposed solution also has been compared with the existing solutions that have been used to segment different types of images.