Automatic Quality Assessment And Perimeter Measurement Of Fetal Abdominal Ultrasound Image

The quality of ultrasound(US)images and the precision of biometric measurement for prenatal US screening are crucial for accurate fetal growth diagnosis.However,manual quality control is a labor intensive process and often impractical in a clinical setting.On the other hand,boundary delineation of anatomies is time-consuming and often leads to large inter-observer variability in clinical practice.To improve the efficiency of examination and alleviate the measurement error caused by improper US scanned plane selection and coarse biometric estimation,a computerized fetal US image quality assessment(FUIQA)and perimeter measurement(FUIPM)scheme are proposed to assist the implementation of fetal abdominal US images quality control and precise biometric measurement in the clinical obstetric examination.The proposed FUIQA is realized with two deep convolutional neural network(DCNN)models,which are denoted as L-CNN and C-CNN,respectively.The L-CNN aims to find the region of interest(ROI)of the fetal abdominal region in the US image.Based on the ROI found by the L-CNN,the C-CNN evaluates the image quality by assessing the goodness of depiction for the key structures of stomach bubble(SB)and umbilical vein(UV).To further boost the performance of the L-CNN,we augment the input sources of the neural network with the local phase features along with the original US data.With comprehensive experiments,it will be illustrated that the proposed FUIQA system is applicable to the fetal US images which gestational ages range from 16 to 40 weeks by setting various parameters and the agreement compared to the subjective ratings from medical doctors is as high as 90%.The proposed FUIPM is a cascaded framework for fully automatic fetal US image segmentation.Firstly,a Rectified Fully Convolutional Network(R-FCN)was utilized to exploit feature extractions from multiple visual scales and distinguish the anatomy with a dense boundary prediction map.To enhance the local spatial consistency and refine the details of the prediction map,we further implant the core R-FCN classifier into an Auto-Context Model and modify the join operator in traditional Auto-Context Model from parallelization to summation.Extensive experimental results show that our FUIPM scheme can bridge severe boundary incompleteness and achieve the best segmentation accuracy when compared with state-of-the-art methods.The proposed FUIQA and FUIPM scheme in this thesis are general,and can be easily extended to the quality assessment and perimeter measurement of other types of fetal US views of fetal face and fetal four cardiac chambers,etc.