![]() | Only 14 pages are availabe for public view |
Abstract This work proposes a new feature set based on applying a polynomial fitting technique on pixels inside the region of interest (ROI). These features will be used to detect the abnormalities in breast tissue by using two classification techniques, the K-nearest neighbor (KNN) and neural networks (NN) and using three databases. Using these features with these two classification algorithms gave very good accuracy in the case of classification between benign and malignant tissues which is 95.24% accuracy using K-NN in case of the Digital Database for Screening Mammography(DDSM) database, and 100% to differentiate normal from abnormal lesion using KNN and NN for the Mammographic Image Analysis Society (MIAS) database. Most of the previous methods used different techniques to decrease the dimensionality of their system, which results in a complicated and computational complexity of the used system. The proposed features in this work are fewer in number, so, there is no need for dimensionality reduction steps. In addition to MIAS and DDSM databases, we used a local data set from “Baheya Foundation for Early Detection & Treatment of Breast Cancer” which is a specialized hospital in diagnosing and treating breast cancer. This database called Contrast Enhancement Spectral Mammography ‘CESM’ which provide a better quality images than the two previous standard databases. CESM has been used to differentiate between normal and abnormal cases to avoid dens overlapping tissues in previous databases. In the meantime, we proposed a method to evaluate the response of tumor to chemotherapy, and the degree of tumor after this treatment where we got an accuracy that approaches about 97.6% on the CESM images. |