Classification of Region of Interest in Mammograms Using Dual Contourlet Transform and Improved KNN

Abstract EN

Goal.

Breast cancer is becoming one of the most common cancers among women.

Early detection can help increase the survival rates.

Feature extraction directly affects diagnosis result.

In this work, a novel feature extraction method based on Dual Contourlet Transform (Dual-CT) is presented, and improved K nearest neighbor (KNN) is employed to improve the classification performance.

Method.

This presented method includes three main sections: firstly, the Region of Interest (ROI) is cropped manually according to gold standard from Mammographic Image Analysis Society (MIAS) database; secondly, the ROIs are decomposed into different resolution levels using Dual-CT, contourlet, and wavelet; a set of texture features are extracted.

Then improved KNN and traditional KNN are implemented for classification.

Experiments are performed on 324 ROIs which include 206 normal cases and 118 abnormal cases; the abnormal cases are composed of 66 benign cases and 52 malignant cases.

Results.

Experimental results prove the validity and superiority of Dual-CT-based feature and improved KNN.

In particular, 94.14% and 95.76% classification accuracy is achieved based on Dual-CT domain.

Moreover, the proposed method is comparable with state-of-the-art methods in terms of accuracy.

Contribution.

Dual-CT-based feature is used for analyzing mammogram and can help improve breast cancer diagnosis accuracy.