An Experimental Analysis Based on Automated Detection of Polycystic Ovary Syndrome on Ultrasound Image using Deep Learning Models

Hormonal imbalances in women of reproductive age, such as PCOS, are a major contributor to female infertility. Acne, menstrual irregularities, excess hair growth, and obesity are all indications of polycystic ovary syndrome. In order to control the symptoms and lessen the health risks connected with PCOS, a quick diagnosis is crucial. Rotterdam criteria are used to make the diagnosis regardless; these include a high androgen hormone level, inability to ovulate, and an ultrasound finding of polycystic ovaries (PCOM). One of the difficult PCOS diagnostic criteria is the diagnosis of PCOM using ovarian ultrasonography that is now performed manually by doctors and radiologists through the counting of follicles and the determination of their thickness in the ovaries. Along with attending to the patient's symptoms, these doctors also need to look for biochemical and clinical indicators of PCOS in order to make a diagnosis. In addition, there is no standard diagnostic procedure or test that doctors use. In order to detect polycistic ovaries from ultrasound pictures of ovarian tissue, we created a Hybrid Deep Learning Model consisting of an Ensemble Inception Model and a LightGBM. The suggested model achieved an accuracy of 98.73% on test pictures, whereas the InceptionV3 achieved an accuracy of 95.34%. This difference was proved by comparing the two models on a number of quantitative effectiveness assessment metrics.