Automatic fetal biometry prediction using a novel deep convolutional network architecture

Purpose Fetal biometric measurements face a number of challenges, including the presence speckle, limited soft-tissue contrast and difficulties in low amniotic fluid. This work proposes convolutional neural network for automatic segmentation measurement fetal parameters, biparietal diameter (BPD), head circumference (HC), abdominal (AC), femur length (FL) from ultrasound images that relies on attention gates incorporated into multi-feature pyramid Unet (MFP-Unet) network. Methods The proposed approach, referred to as Attention MFP-Unet, learns extract/detect salient regions automatically be treated object interest via gates. After determining type anatomical structure image using network, Niblack's thresholding technique was applied pre-processing algorithm abdomen identification, whereas novel used extraction. A publicly-available dataset (HC18 grand-challenge) clinical data 1334 subjects were utilized training evaluation MFP-Unet algorithm. Results Dice similarity coefficient (DSC), hausdorff distance (HD), percentage good contours, conformity coefficient, average perpendicular (APD) employed quantitative anatomy segmentation. In addition, correlation analysis, evaluate accuracy biometry predictions. achieved 0.98, 1.14 mm, 100%, 0.95, 0.2 mm DSC, HD, conformity, APD, respectively. Conclusions Quantitative demonstrated superior performance compared state-of-the-art approaches commonly parameters.