2011 International Conference on Image Information Processing (ICIIP 2011) Proceedings of the 2011 International Conference on Image Information Processing (ICIIP 2011) 978-1-61284-861-7/11/$26.00 ©2011 IEEE Automated Ovarian Follicle Recognition for Polycystic Ovary Syndrome Palak Mehrotra , Chandan Chakraborty School of Medical Science and Technology Indian Institute of Technology Kharagpur, West Bengal palakmehrotra@gmail.com chandanc@smst.iitkgp.ernet.in Biswanath Ghoshdastidar, Sudarshan Ghoshdastidar, Kakoli Ghoshdastidar G D Institute for Fertility Research Kolkata, West Bengal bgdastidar@gmail.com sudarshan.ivf@gmail.com Abstract— Polycystic Ovary Syndrome (PCOS) is a complex endocrine disorder affecting many women in the pubertal as well as reproductive age groups with profound adverse affects such as obesity, infertility, cardiovascular disease and diabetes mellitus . Diagnosis of the condition is by clinical, biochemical and imaging parameters. The principle feature on ultrasound is the presence of polycystic ovaries with peripheral arranged cysts and dense stroma. During ultrasound evaluation due to overlapping of the follicles as well as inherent noise of the equipment delineating, making this characteristic appearance may sometimes become challenging, making diagnosis time consuming. Moreover the interpretation would vary considerably from one operator to another as it is largely an experience dependent procedure. In this paper an automated scheme for the detection of this pathognomonic pattern and arrangement of follicles is proposed to overcome this problem. Firstly the input ultrasound image was preprocessed by multiscale morphological approach for contrast enhancement. Then a scanline thresholding is used to extract the contours of the follicles. The results are compared with the results obtained by manual selection to verify the effectivity of scheme. Keywords- Polycystic Ovary Syndrome; Multiscle Morphology; Scanline Thresholding I. INTRODUCTION The Polycystic Ovary Syndrome (PCOS) is a complex metabolic disorder affecting 5-10% women in the reproductive age [1, 2]. The exact aetiopathogenesis of PCOS is yet to be fully understood, through genetic and environmental factors have been implicated. The syndrome is characterized by the classical traits of menstrual irregularity, hyperandrogenism and polycystic ovaries on ultrasound [3, 4]. Associated health risks are considerable. A significant proportion of these women suffer from cardiovascular disease and Type II diabetes mellitus in the long term. Moreover, most of these women will suffer from anovulatory infertility. Early detection and treatment significantly improve the chances of conception as well as offer better long term health prognosis especially as often the disease is manifested early in puberty and worsen with the age and lack of treatment. The Rotterdam consensus [5] is currently the most important criteria for diagnosing this condition According to it a patient may be diagnosed as suffering from PCOS as long a she has any two of the following three criteria: (i) Chronic Anovulation- clinically manifested as irregular/ infrequent menstrual cycles, (ii) Hyperandrogenism- clinically evident as acne, hirsutism and elevated serum enzymes, (iii) PCOS ovaries on ultrasound. PCOS ovaries have a characteristic appearance on ultrasound imaging showing the presence of multiple cysts, typically 12 or more measuring between 2-9 mm in a peripheral distribution along the border of the ovary , classically described as the “necklace formation” [6,7]. Often they may be distributed throughout the ovary as well. It is thought that these cysts are developed due to the hormonal imbalance which characterize these patients is ambiguous and not yet fully clear. Moreover, the ovarian volume in such patients is typically increased over 10 cm 3 . During ultrasonographic evaluation of the ovary in a suspected PCOS patient, overlapping of the follicles as well as noise of the equipment may make diagnosis of polycystic ovaries time consuming and tedious. Moreover, results would vary considerably from one operator to another as it is largely an experience dependent procedure. Thus we tried to develop a model for automated detection of the polycystic ovary on ultrasound. The literatures reported on the computer assisted approach for follicle detection are very few. In [8] Potocnik and Zazula based on optimal thresholding and then further upgraded by using active contour [9] for the segmentation of follicles. In [10] Cigale and Zazula implemented the neural network approach for the segmentation. Anthony Krivanek and Milan Sonka [11] segmented the wall of the follicles based on watershed segmentation and knowledge-based graph search algorithm. The ultrasound images have noise which lead to poor segmentation as the edges are not visible. In [12] Hiremath and Tegnoor used the contourlet method for the denoising of the ultrasound image. In this paper firstly multiscale morphology was applied for the noise suppression and enhancing contrast which is the