Characterisation of Breast Cancer Lesions using Image Processing Based Technique

Introduction: Characterisation of mammographs deliberates as influential approaches in cataloguing of breast tissues and tumour. In breast, unravelling of nearby tissues in the mammographs is one of the tough processing procedures. The existence of speckle noise in these mammographs boundaries makes the pathology analysis more difficult. Aim: To characterise breast cancer lesions using different image processing algorithms in order to improve the mammographs and increase their diagnostic value. Materials and Methods: This retrospective study aims to locate structures and lesions in breast images. The algorithms use the noise speckles deletion, augmentation and subdivision of the breast tissue and the background in mammographs. More precisely, it aims to ascribe a label to pixels within the mammographs that have the same graphic characteristics. The segmented images were associated with the binary image mask to the original mammograph. The Root-Mean-Square Error (RMSE) and Peak Signal-to-Noise Ratio (PSNR) were studied in images database. Both percentage match between ground truth and segmentation results were calculated. Results: Percentage match measure of watershed algorithm was 96.60 (p<0.05) and Corresponding Ratio (CR) was 0.019 (p<0.05). The edge detection gave good and clear visualisation of the processed images that increased the diagnostic value of them. Conclusion: The edge detection and water-marker technique are able to identify the breast lesions precisely and improves radiological analysis and diagnosis.