Article Preview
Top1. Introduction
Uterine fibroids, also referred as uterine leimyomata are the cancerous tumors found in the women of the reproductive age group (Hassani, 1975; Yao et al., 2006; Acien, 1996)
Histologically these tumors are composed of bundles of smooth muscle cells found with in the walls of the uterus. The presence of hormones such as estrogen, progesterone merely influences the development of fibroids which are classified into the categories, submucosal, intramucosal and suberosal according to their position within the uterus (Acien, 1996; Mukhopadhaya, et al., 2007; Siskin, et al., 2006; Walker et al., 2007). It is well known that the symptoms such as, heavy menstrual bleeding, irregular vaginal bleeding are the clear indicators of the presence of fibroids. The challenging issue for the clinical specialists is to assess the uterus images to decide whether the tumor is benign or malignant. The less calcified benign tumor can be surgically removed whereas the malignant tumor can cause infertility as well as repeated miscarriage. In the current clinical practice, only qualitative visual procedure is adapted and there is no computer aided diagnostic tool exists to support the clinicians. Earlier literature reveals that only few study have been reported for automated uterine fibroid detection (Malarkhodi & Wahida Banu, 2012; Ratha et al., 2010; Sriraam et al., 2010). Malarkodi and Wahida Banu proposed an automated segmentation technique using local phase based level set approach (Malarkhodi & Wahida Banu, 2012). Discrete wavelet transform and histogram equalization was applied to remove the speckle noise. Shape based features, such as, total area, perimeter, diameter and eccentricity were extracted. The study was carried with only five images. Ratha and Ramar have shown a knowledge based approach for detection of fibroids (Ratha et al., 2010). A modified morphological image cleaning (MMIC) algorithm was applied to remove the speckle noise and a heuristic rule based approach was introduced for segmentation. Shape based features such as diameter, area and compactness were applied and empirical evaluation method was adopted to determine the classification accuracy . Sriraam et al. (2010) employed wavelet packet features with backpropogation neural network classifier and have shown a classification accuracy of 95.1%. In order to achieve better pattern recognition accuracy, this research study suggests the development of computer aided diagnostic tool for automated detection of uterine fibroids using Gabor wavelet based features with neural network classifier. Section 2 highlights the proposed automated detection scheme. In section 3, gray level intensity and morphological operator analysis are described to see the distinct variation between the normal and the fibroid uterus images. Section 4 discusses the feature extraction and classification schemes. The performance evaluation of the proposed scheme is shown in section 5 and a detailed discussion is reported in Section 6. Section 7 highlights the important concluding remarks.