Research On Automatic Segmentation And Classification Index Of PCOS Ultrasound Image

With the continuous advancement of science and technology and the continuous updating of medical technology,various diseases that could not be completely cured or even incurable in the past have gradually been perfectly treated.However,there are still some diseases that are still incurable and there are no exact diagnostic criteria.Polycystic ovary syndrome(PCOS)is a serious disease that affects women's health.It mainly causes women's endocrine disorders,leading to hormonal imbalances,even abnormal menstruation,obesity and infertility and also increases the probability of diabetes and hypertension.Therefore,it is very important to diagnose and prevent PCOS in advance to reduce the impact of disease.The main indicator for diagnosis of the disease in modern medicine is the number,size and distribution of follicles in ovarian ultrasound images.In this paper,the feature extraction of this index will be carried out from the perspective of image segmentation.The specific research work is as follows:(1)In machine learning,an image segmentation method(PSO-FCM-Watershed)combining particle swarm optimization with fuzzy C-means clustering(PSO-FCM)algorithm and Marker control watershed method is proposed.The image is preprocessed firstly,then the image is clustered using PSO-FCM to enhance the contrast of the image,and the follicles boundary is segmented in conjunction with the watershed.Compared with the traditional segmentation method,the PSO-FCM-Watershed proposed in this paper has improved the average crossover ratio(MIOU)and the segmentation accuracy by nearly 4% and 14%.At the same time,it has been verified that the proposed method has a good segmentation effect.(2)In terms of deep learning,a deep learning method(H-U-net)combining histogram equalization and U-net is proposed.The first step in this paper is mainly the denoising and histogram equalization processing,and then the method is combined with U-net to train model and segment image.The method proposed in this paper has improved the evaluation index by 17% and 15% compared with U-net.In view of the small amount of medical data,the proposed method can train better segmentation models,which is superior to the common U-net method.(3)Comparative experiments were carried out on the two methods proposed above to verify the effectiveness of the proposed method.In terms of machine learning,PSOFCM-Watershed is compared with traditional methods firstly.The accuracy rate is much higher than the traditional method,and it is better than U-net.The superiority of the method can also be confirmed.On the other hand,H-U-net also shows better results than ordinary U-net,which provides a new idea for deep learning to train the model.At the same time,we also discuss the classification indicators of normal people and PCOS patients.According to the experimental results,the conclusions about the classification indicators obtained in this paper are consistent with the currently accepted classification diagnostic indicators to some extent,and provide a theoretical basis for the later research.