Preprocessing of Breast Cancer Digital Mammograms for AI-Based Detection and Classification Models
جاري التحميل...
التاريخ
عنوان الدورية
ردمد الدورية
عنوان المجلد
الناشر
Journal of Systems Science and Systems Engineering
خلاصة
Breast cancer detection images are the gold standard in the diagnosis and prognosis of the disease. For early diagnosis, the digital mammogram has become the most preferred screening procedure. Mammography has several artifacts that have a deleterious impact on the detection of breast cancer. As a result, eliminating artifacts and improving image quality is necessary in the Computer Aided Diagnosis (CAD) systems. The accuracy and efficiency of the CAD are improved by giving precise Regions of Interest (ROI). ExtractingROI is difficult, since the existence of pectoral muscles affects the detection of abnormalities. The proposed system aims to enhance image quality, facilitate feature extraction, and improve diagnostic accuracy in medical imaging applications. It demonstrates the integration of various techniques to achieve optimal results for clinical analysis and decision-making. The system consists of four key stages. First, remove the image background and eliminate any unwanted noise. Second, segment pectoral muscle using a seed point selection technique based on image orientation. This step is important to identify and delineate the pectoral region within the image. Third, image enhancement techniques to improve the visual quality of the processed image using the Wiener filter for noise reduction and sharpening. Fourth, histogram equalization and intensity refined using a CLAHE filter to enhance local contrast, followed by an intensity filter to refine the image quality. Two different benchmark datasets DDSM, and breast mammography (tomosynthesis) are used to validate the generality and efficiency of the proposed method. The findings of experimental results on the entire database demonstrate that the overall performance of the proposed method evaluated achieves higher values of mean and STD, which indicates the efficiency of the proposed method. It also achieves an average SSIM of 94%, indicating high image similarity even after processing. Additionally, breast and muscle for the entire database are successfully extracted, removing background, achieving 100%. The correctness percentage is 99.04%, while the completeness achieves is 97.49%.