Equation 1 is used to convert the measured frequency shift to velocity. As equation 1 clearly indicates, it is important to have precise knowledge of the angle between the ultrasound beam and the flow direction in order to calculate the velocity. This angle is commonly referred to as the “Doppler angle”. In current clinical practice, the ultrasound operator (sonographer) would manually specify the angle between the Doppler ultrasound beam and the vessel orientation while performing the ultrasound exam in order to ensure that the correct flow velocities are estimated. This is clinically referred to as angle correction, and is one of multiple steps in the clinical workflow for Spectral Doppler. Angle correction refers to adjustment of the Doppler angle and is used to calibrate the velocity scale for the angle between the US beam and the blood flow being measured. Commercial ultrasound scanners have a dedicated control on the user interface that allows the sonographer to specify the angle correction. Nevertheless, it is widely recognized that incorrect angle correction is a leading cause of error in blood velocity measurements performed using Doppler ultrasound. In fact, it is noted in the top three amongst Top Ten Doppler Errors and Artifacts [Pegasus lectures]. Dejong13 from the Intersocietal Com-mission for the Accreditation of the Vascular Laboratories (ICAVL) commission reported that as high as 35% of the applications for accreditation received by the ICAVL demonstrate improper angle correction techniques, making angle correction issues one of the most common causes for delayed decisions.

Recent developments in computer vision has led to a number of automated approaches for Doppler angle estimation.  \citet{Hirsch2006} presented a multi-scale approach for estimating the vessel's flow direction by principal component analysis. More recently, \citet{Saad_2008} described a computer vision approach to automate the Doppler angle estimation. The approach starts with the segmentation of blood vessels in ultrasound color Doppler images. The segmentation step is followed by an estimation technique for the Doppler angle(\(\theta\)) based on a skeleton representation of the segmented vessel. Statistical regression analysis showed strong agreement between the manual and automated methods. They further hypothesized that the automation of the Doppler angle will enhance the workflow of the ultrasound Doppler exam and achieve more standardized clinical outcome.

 

The above automated techniques are based on computer vision and image analysis techniques. Saad et. al applied techniques to segment the vessel from the image in order to determine the vessel orientation. The input images used were duplex color flow images. The color flow information was used to identify the vessel regions. A potential drawback of such an approach is that artifacts in the color flow images can affect subsequent steps in the algorithm. In this paper, we propose a novel automated Doppler angle estimation technique based on a deep-learning framework. The method is designed to operate directly with B-mode (grayscale) images without using any color information. Moreover, the input data to the deep learning network is the raw image without any pre-processing or use of segmentation. We hypothesize that the image features necessary for accurate angle estimation are discovered by the deep learning algorithm itself. The longer term motivation is that such a general approach makes the technique more broadly applicable especially when the image quality and appearance is known to vary significantly between ultrasound scanners due to the use of proprietary front-end hardware configurations and sophisticated post-processing software.