Classification of microcalcification in mammogram images using Enhanced Support Vector Machine (ESVM) / Muhammad Akmal Firdaus... [et al.]

• Main Authors: Firdaus, Muhammad Akmal, Shadan, Siti Madinah, Mohd Faudzai, Nur Syahizah
• Format: Student Project
• Language: English
• Published: 2019
• Subjects: Study and teaching; Mathematical statistics. Probabilities; Analytical methods used in the solution of physical problems
• Online Access: http://ir.uitm.edu.my/id/eprint/36987/1/36987.PDF; http://ir.uitm.edu.my/id/eprint/36987/

Support Vector Machine (SVM) is a supervised machine learning algorithm with the ability to build a classification model from a labeled dataset. SVM has been broadly used in image classification of medical imaging such as mammogram images for breast cancer detection due to its higher classification precision, higher prediction accuracy, better generalization capability and better overall performance. Breast cancer detection is critically dependent on early detection and accurate diagnosis. However, the existed histopathological classification of breast cancer has clinical utility that is limited due to insufficient prognostic and predictive power. The diagnostic decisions by experienced physicians can be increased by an effective medical decision support system. Therefore, this research concentrates on Enhanced Support Vector Machine (ESVM) that combines Discrete Wavelet Transform (DWT) and Principal Component Analysis (PCA) for better data classification accuracy. DWT is used for the extraction of statistical and textures features of mammogram image. Meanwhile, the employment of PCA is to reduce the dimensionality of the datasets to avoid overfitting thus better accuracy can be achieved. The ESVM focuses on classification of microcalcification in 30 mammogram images obtained from the National Cancer Society Malaysia (NCSM). The performance of the ESVM method is measured in term of classification rate (CR), specificity (SP), sensitivity (SY) and the accuracy based on Receiving Operating Characteristics (ROC) curve obtained by comparing the images classification results of ESVM and expert findings. The result revealed that the classification rate is 90% while for the accuracy based on the ROC Curve is 0.9375. Based on the result obtained, it is proven that ESVM is excellent in classifying the microcalcification in mammogram images and act as an assistance for radiologists that provides "second judgment" on medical image readings.