Classification of breast tumors using textural features
| Main Authors: | Ali Abbasian Ardekani, Akbar Gharbali, Afshin Mohammadi |
|---|---|
| Format: | Article |
| Bahasa: | eng |
| Terbitan: |
, 2014
|
| Online Access: |
https://zenodo.org/record/2659905 |
Daftar Isi:
- Abstract— Breast cancer is a major public health problem in women from developed and developing countries. Early detection and treatment of breast cancer increase the cure rate and provide optimal treatment. The purpose of this study was to evaluated computer aided diagnostic (CAD) methods with texture analysis (TA) To improve radiologists accurate in classification breast tumors as a malignant or benign. The database consist of US images of the breast patients (18 benign and 15 malignant breast cancers). Mazda software (version 4.6) was used for TA of breast cancers by loading one image per patient and generating 80 region of interests (ROIs) (48 benign and 32 malignant). We extract 300 statistical texture features as a descriptor for each selected ROI. Then features were eliminated to 10 best and most effective features by Minimum error probability combined with average correlation coefficients (POE+ACC). 1%-99% normalization and two texture analysis methods: Principle Component Analysis (PCA), Linear Discriminant Analysis (LDA) were tested. The resulting features classified by first Nearest-Neighbor (1-NN) classification. Receiver operation characteristic (ROC) curve analysis used for evaluation applied TA methods performance by calculating area under ROC curve (A_z), positive predictive value (PPV), Negative predictive value (NPV), sensitivity (SN), specificity (SP), and overall accuracy (ACC). Best results are Obtained via NDA to discrimination Benign from malignant breast tumors with sensitivity of 91.42%, specificity of 91.07% and accuracy of 91.2% and The area under the ROC curve (A_Z) of 0.124. The results showed that TA is a reliable method and has a potential to use classification benign tumor and malignant in breast US images. The methods developed in this study could be used as a part of CAD system to reduce biopsy and improve the accurate of radiologist for differentiate between benign and malignant tumors.