Complete Fetal Head Compounding from Multi-view 3D Ultrasound

摘要: Ultrasound (US) images suffer from artefacts which limit its diagnostic value, notably acoustic shadow. Shadows are dependent on probe orientation, with each view giving a distinct, partial of the anatomy. In this work, we fuse partially imaged fetal head anatomy, acquired numerous views, into single coherent compounding full Firstly, stream freehand 3D US is acquired, capturing as many different views possible. The anatomy at time-point then independently aligned to canonical pose using an iterative spatial transformer network (iSTN), making our approach robust fast and motion. Secondly, fused by averaging only best (most salient) features all images, producing more detailed compounding. Finally, iteratively refined groupwise registration approach. We evaluate quantitatively qualitatively, comparing it average individual frames. also alignment accuracy two physically attached probes, that capture separate simultaneously, providing ground-truth. Lastly, demonstrate potential clinical impact method for assessing cranial, facial external ear abnormalities, automated atlas-based masking volume rendering.