Machine learning approaches in medical image analysis of PCOS

Polycystic Ovary Syndrome (PCOS) is a complex hormonal disorder which is associated with diverse symptoms such as irregular menstrual cycles, obesity, acne issues, hirsutism, alopecia, and excess secretion of androgen in women. These symptoms may vary from patient to patient and therefore it is important to get it diagnosed at an early stage as it may lead to serious health conditions like heart diseases, diabetes, endometrial cancer etc. Diagnosis of PCOS is performed using physical examination, laboratory tests and ultrasound imaging. Ultrasound is used to detect follicles or immature eggs present inside an ovary. Ovaries with PCOS can be easily detected using ultrasound where trans vaginal ultrasound is considered to be the most efficient technique as it provides an internal view of the ovaries especially in patients suffering from obesity. The ultrasound report consist of details like number of follicles present, and size of ovaries, luminal thickness etc., Manual detection is costly, and lengthy and may be prone to physician's error due to shortage of health staff in developing countries like India and lead to higher false positive rates. Therefore, an automated system for PCOS segmentation and classification can help in escalating this task. This paper reviews existing state-of-the-art works where classification and segmentation have been used for automated detection of PCOS. Various machine learning, deep learning and transfer learning techniques are discussed. It was observed that this research field needs technological advancements through release of open-source datasets, and robust automatic pipelines for fruitful detection of PCOS in women.