Comparison of features detection methods in classifying types of breast tissue

Breast cancer is now the leading cause of cancer death in women. Computer-Aided diagnosis (CADx) has proven effective in assisting medical experts in making an early diagnosis, increasing the probability of recovery. Machine learning is used in Computer-Aided diagnostics to predict whether a patient has benign or malignant cancer. Feature selection algorithms can be utilised to produce more accurate results. Our problem statement is to fill the gap by classify the multiclass type of classification for the breast tissue using SURF as a feature detection approach that distinguishes from other researchers and attempt to compare the performance for both FAST and SURF feature detection. We discovered that Speed Up Robust Feature (SURF) required more computational time than Features from Accelerated and Segments Test (FAST) to complete the feature detection phase. Thus, this study compared and measured the performance of SURF and FAST descriptors for dense, fatty, and glandular tissue. In this report, the digital mammogram image is pre-processed first, this is followed by segmented the pre-processed images. Then, the feature detected, and feature extracted phase will be carried out. Next, the classification phase is conducted using Fine K-nearest Neighbours (KNN). Finally, the model is evaluated. As a result, the SURF method shows the best accuracy for training, while FAST gives the better accuracy for testing. Thus, using the appropriate threshold can help distinguish between tissue and background photos. Both the segmentation phase and the extracted features phase are important to the effective performance of the CADx.