Analysis of margin sharpness for breast nodule classification on ultrasound images

Breast cancer has the highest prevalence, incidence and mortality for females in worldwide and no exception in Indonesia. Ultrasound is a recommended modality for diagnosing breast cancer through ultrasound images. However, misdiagnosis might still occurs which is caused by human factors. Margin of...

Full description

Saved in:
Bibliographic Details
Published in2017 9th International Conference on Information Technology and Electrical Engineering (ICITEE) pp. 1 - 5
Main Authors Adi Nugroho, Hanung, Triyani, Yuli, Rahmawaty, Made, Ardiyanto, Igi
Format Conference Proceeding
LanguageEnglish
Published IEEE 01.10.2017
Subjects
Breast cancer
breast nodule
Feature extraction
Image edge detection
Image segmentation
margin sharpness
neutrosophic
Ultrasonic imaging
ultrasound
watershed
Online AccessGet full text

Cover

More Information
Summary:Breast cancer has the highest prevalence, incidence and mortality for females in worldwide and no exception in Indonesia. Ultrasound is a recommended modality for diagnosing breast cancer through ultrasound images. However, misdiagnosis might still occurs which is caused by human factors. Margin of breast nodule is one of the malignancy characteristics based on BIRADS. This research proposes a computer aided diagnosis (CADx)-based method for classifying breast nodules in ultrasound images based on margin characteristics. In practice, CADx is used as a second opinion in interpreting ultrasound images in order to obtain more accurate diagnosis results. The proposed approach consists of adaptive median filter for marker removal, pre-processing with normalisation and speckle reduction anisotropic diffusion (SRAD) filter followed by neutrosophic and watershed methods for segmentation process, features extraction and feature selection. A total of ten selected features including of texture, geometry and margin sharpness features are then classified by using multi-layer perceptron (MLP). This study uses 102 breast ultrasound nodule images with 57 non-circumscribed and 45 circumscribed margins. The performance of proposed approach achieves the accuracy of 95.10%, sensitivity of 93.33%, specificity of 96.49%, PPV of 95.45%, NPV of 94.83%, Kappa of 0.9004 and area under curve (AUC) of 0.989. These promising results indicate that the proposed approach successfully classifies breast nodule based on margin characteristics has a potential for assisting the radiologists in interpreting breast ultrasound images.
DOI:10.1109/ICITEED.2017.8250442