Research On Segmentation Methods Of Fetal Corpus Callosum Based On Deep Learning

Abnormal corpus callosum is a congenital developmental abnormality in fetal central nervous system malformation,which refers to partial or complete loss of the corpus callosum during fetal development.In general,the fetus with abnormal corpus callosum has no obvious external symptoms and is difficult to be detected before the onset of the disease.However,there may be visual impairment,epilepsy,mental retardation,mood disorders and other diseases in the future.In clinical practice,using ultrasound technology to detect abnormalities in the corpus callosum of the fetus often depends on manual diagnosis.Doctors rely on their own work experience to find the corpus callosum area of the fetal brain,and then generate and analyze ultrasound images,which is a huge workload.Therefore,in order to reduce the burden of doctors and improve the accuracy of diagnosis of fetal corpus callosum abnormalities,the task of fetal corpus callosum segmentation based on ultrasound images becomes indispensable.Based on the above realistic background,we study the method of fetal corpus callosum segmentation base on deep learning from the following two aspects:Segmentation of fetal corpus callosum based on codec network structure.Firstly,according to the characteristics of ultrasound images of fetal corpus callosum,we develops a data collection scheme and labeling specifications,and uses gamma correction,histogram equalization and other methods to enhance data and inhibit the occurrence of over-fitting phenomenon.Secondly,we introduces the codec network structure used in detail,and focuses on the modification based on the original method.Finally,in order to find the optimal segmentation method for ultrasonic images of fetal corpus callosum,we conducts comparative experiments in four directions of image data enhancement,optimizer,loss function and codecs network respectively,and finds out appropriate model structures to achieve high-precision segmentation of fetal corpus callosum.Segmentation of fetal corpus callosum based on attentional mechanism.Firstly,we introduces the current mainstream attention mechanism in detail,including multi-headed attention mechanism,bilateral attention mechanism and window attention mechanism.Secondly,we combines these three attention mechanisms with the actual situation of the task of dividing fetal corpus callosum,builds a network model for dividing fetal corpus callosum,and proposes a new fusion attention mechanism for dividing fetal corpus callosum.Finally,in order to verify the performance of the three models in the task of dividing fetal corpus callosum,detailed comparative experiments have been conducted in this thesis.The results show that the fusion attention method proposed in this thesis can effectively improve the performance of fetal corpus callosum segmentation.Finally,this thesis collects the fetal corpus callosum ultrasound image dataset,and realizes the high-precision segmentation of the fetal corpus callosum ultrasound image by using the fusion attention mechanism algorithm and the window attention mechanism algorithm through the deep learning technology.The m Io U values reach 0.787 and 0.788 respectively,which can be used to reduce the diagnostic workload of doctors and improve the diagnostic rate of fetal corpus callosum abnormalities.