Shape Estimation of Flexible Probe using Spatial-Attention FlexShapeNet

Ultrasound imaging with flexible probes requires the knowledge of element array geometry, which is variable and unknown. Here, we propose a spatial-attention shape estimation model, named FlexShapeNet, to estimate accurate element array geometry of flexible probes from radio-frequency (RF) data. A spatial attention (SA) module is introduced to capture deformation information resulting from variations in the shape of the flexible probe, which effectively reduces the significant disparities in shape estimation accuracy among different imaging targets. The delay-and-sum (DAS) beamformer is then used to reconstruct ultrasound images with estimated shape and the raw RF data. Experimental results demonstrate that FlexShapeNet achieves the highest shape estimation accuracy for flexible probes without the need for multiple iterations of beamforming.