Design and Analysis of an Algorithm for Breast Tumor Segmentation in Mammogram and Ultrasound Images

The breast cancer is having high mortality rate among ladies. The current trend of identifying the cancerous tumor is by medical image processing such as mammogram and ultrasound. The heart of medical image processing lies in the segmentation of the tumor in the mammogram. Still the conventional method of segmentation faces many dynamic challenges due to various noises such as Gaussian, Pepper & Salt, and Speckle noise hence eliminating such noise and segmenting tumor with high precision from the ultrasound and mammogram images is the goal. The task of finding the suitable segmentation algorithm for the segmentation of different medical images with a high accuracy plays a vital role and is another challenge. Also the current segmentation algorithm misguides the actual feature extraction of the tumor and also leads to high mortality rate in ladies. In this work the medical images are enhanced to avoid the various noises using modified Gabor filter and estimated the quality of the mammograms for the segmentation with the metrics MSE (Mean Square Error) and PSNR (Poisson Signal to Noise Ratio) of the image. Several segmentation algorithms like Otsu, SRM (Statistical Region Merging), Region growing& merging and FCM (Fuzzy C means clustering) are applied on images. Along with that five edges based segmentation algorithms like Canny, Sobel, LoG(Laplacian of Gaussian), Prewitt and Roberts are also applied and their performance has been measured with respect to gold standard images of the Berkeley Database. In this research work region growing and merging and FCM and Otsu had been adopted for tumor segmentation and region growing and merging has performed better for breast cancer tissue segmentation in the medical images. The performance of the Region growing and merging, FCM and Otsu segmentation has been measured by the metrics like F-score with the value 0.9673, 0.9573 and 0.9489 respectively. Hence these three algorithms can be adopted for the better segmentation of the breast image.