Speckle Reduction for Improved Classification

Project Github Repository

Students: Inbal Aharoni, Shani Israelov, Supervised by: Shira Nemirovsky-Rotman

In this project, different speckle reduction techniques have been studied and implemented in software,and then tested on breast images in order to examine the impact on lesions classification as malignant or benign.

Speckle pattern tends to obscure edges and reduce the image contrast and therefore is interrupting the radiologists in the medical diagnosis.

The first method is OBLMN Algorithm which we didn’t implemented ourselves and the second is Adaptive Wavelet Thresholding which we implemented in MATLAB.

We examine the impact on lesions classification in two ways. The first one is quantitative Features’ Extraction. The features divided into morphological and textural features that characterize the mass.The second way to examine the impact is by radiologist professional opinion. Finally, we made conclusions.