Force Coupling Construction and Signal Processing
A force-sensing casing for the Philips XL14-3 xMATRIX ultrasound probe was assembled from 3-D printed from ASA plastic shells (MakerBot Industries, LLC, New York City, New York, United States) and an aluminum frame. Within the casing are an LSB205 S-beam load cell (FUTEK Advanced Sensor Technology, Inc.; Irvine, California, United States) to measure force applied to the ultrasound probe imaging surface and an ADXL335 accelerometer (Analog Devices, Inc., Wilmington, Massachusetts, United States) to provide probe orientation information and account for its effects on force measurement. The load cell signal is amplified by the IAA100 differential amplifier (FUTEK Advanced Sensor Technology, Inc.; Irvine, California, United States), which digitizes, along with the accelerometer signal, into the USB-6001 DAQ (National Instruments; Austin, Texas, United States). More detail on the force coupling is provided in the Supplementary Information.
Data Acquisition
Ultrasound data is observed in real time and recorded on the EPIQ 7C Ultrasound System (Philips; Amsterdam, Netherlands). Force data is observed in real time and recorded with LabVIEW 2016 software (National Instruments; Austin, Texas, United States). Given a new orthostatic position of the subject, the orthogonal incidence angle of the force-coupled ultrasound probe with the long axis of the IJV is noted on the LabVIEW interface in terms of yaw and pitch. This angle of incidence is to be maintained throughout each compression in the given orthostatic position. When recording, three quick compressions are applied to IJV. Then, the user slowly and linearly ramps up force to slightly more than what is necessary for complete occlusion of the IJV. Then, one quick compression is applied. These steps are taken to provide recognizable artifacts in the ultrasound images and the force signal for synchronization later. When the Valsalva maneuver is performed by the subject, the only difference is that the subject does it during only the force sweep using the CR410 digital manometer (EHDIS Car Accessory Co. Ltd.; Guangdong, China) to read the airway pressure applied during the Valsalva. More detail on data acquisition is provided in Supplemental Information (figure on synchronization compressions, manometer picture).
Jugular Venous Pulsation Measurement
A subject is slowly tilted downward on a tilt table starting from 60 degrees until the angle where IJV pulsations are visible at the base of the neck. This angle is recorded with an iPhone8 angle sensor app (need more info) and converted to a hydrostatic pressure by assuming 10 cm between the right atrium and the base of the neck and converting cmH2O to mmHg.
Synchronization of Force and Ultrasound
Similar to the synchronization of force and ultrasound images for carotid artery force-coupled ultrasound images23, we derive an optical flow based position signal from the ultrasound images and use peak detection to find the three quick compressions preceding the force sweep, the force sweep maximum, and the one quick compression succeeding the force sweep. The pairs of first and last peaks are aligned and the three middle peak pairs (two quick compressions and force sweep) are checked to be within the error threshold of an absolute sum of about 0.7 seconds disagreement. Then a force is assigned to each ultrasound image. All synchronization data processing is done in MATLAB 2021b (The MathWorks, Inc.; Natick, Massachusetts, United States).
Detection of Internal Jugular Vein
In order to initiate segmentation of the IJV in the force-coupled ultrasound images, the IJV must be detected and a seed point must be provided near the center of the IJV. The IJV is primarily detected via Faster RCNN an object detector neural network, trained on 3000 IJV ultrasound images in a similar manner to the automatic detection of the carotid artery23. To avoid the small risk of failure to detect the IJV, the user can click in the IJV when presented with a synchronized force-coupled ultrasound image to initiate segmentation. More detail is provided in the Supplemental Information (Faster RCNN training and results). All IJV detection is done in MATLAB 2021b (The MathWorks, Inc.; Natick, Massachusetts, United States).
Segmentation of Internal Jugular Vein
When the IJV is not near collapse, region growing from a seed point is followed by a radial line intensity difference maximization, intensity difference and shape optimization, and finally a 2000-point third-order interpolation. The differences from previous work on carotid segmentation are 32 radial lines are drawn after region growing instead of 16 and the 2000-point interpolation replaces the ellipse fit at the end of segmentation23. When the IJV is near collapse (area of previous image IJV is less than 5 mm2), only region growing is occurs because IJV wall pixels tend to be high-intensity when near collapse such that the region does not grow outside of the walls. Area measurement occurs after the segmentation for a frame is complete by creating a binary image differentiating what is outside the segmentation boundary from what is inside and counting the pixels which are inside.
Inverse Finite Element Modeling
Creation of the three-dimensional forward finite element model was done using COMSOL Multiphysics software version 5.6 (COMSOL Inc.; Burlington, Massachusetts, United States). Running an inverse optimization involved using COMSOL’s LiveLink functionality with MATLAB (The MathWorks, Inc.; Natick, Massachusetts, United States), allowing MATLAB to run and adjust COMSOL models.