First In-Vivo demonstration of 1000fps High Speed Coronary Angiography (HSCA) in a swine animal model

Abstract

High-speed-angiography (HSA) 1000 fps imaging was successfully used previously to visualize contrast media/blood flow in neurovascular anatomies. In this work we explore its usage in cardiovascular anatomies in a swine animal model.

A 5 French catheter was guided into the right coronary artery of a swine, followed by the injection of iodine contrast through a computer-controlled injector at a controlled rate of 40 (ml/min). The injection process was captured using high-speed angiography at a rate of 1000 fps. The noise in the images was reduced using a custom built machine-learning model consisting of Long Short-term memory networks. From the noise reduced images, velocity profiles of contrast/blood flow through the artery was calculated using Horn-Schunck optical flow (OF) method.

From the high-speed coronary angiography (HSCA) images, the bolus of contrast could be visually tracked with ease as it traversed from the catheter tip through the artery. The imaging technique’s high temporal resolution effectively minimized motion artifacts resulting from the heart’s activity. The OF results of the contrast injection show velocities in the artery ranging from 20 – 40 cm/s.

The results demonstrate the potential of 1000 fps HSCA in cardiovascular imaging. The combined high spatial and temporal resolution offered by this technique allows for the derivation of velocity profiles throughout the artery’s structure, including regions distal and proximal to stenoses. This information can potentially be used to determine the need for stenoses treatment. Further investigations are warranted to expand our understanding of the applications of HSCA in cardiovascular research and clinical practice.

1. INTRODUCTION

High Speed 1000 fps Angiography is a novel interventional imaging technique previously demonstrated with invitro models representing neurovascular anatomy¹. Using this imaging technique, details of contrast flow such as flow vortices were easily visualized. From these images blood flow velocities were calculated using algorithms such as optical flow² and X-PIV.³ In this work, we focus on high-speed coronary artery angiography (HSCA) for an in vivo cardiovascular artery swine animal model. The overall goal is for HSCA imaging to aid in diagnosis and treatment of cardiovascular diseases such as Coronary Artery Disease (CAD).

2. METHODS AND MATERIALS

2.1. Animal Model and intervention setup

A 5 French guide catheter was inserted into the femoral artery of a 150 lb swine, kept under general anesthesia. The heart rate of the animal was at an average of 88 bpm. The animal was medicated with Heparin as part of standard model care for interventional procedures. No additional medication was given to perform HSCA. Under regular flat panel detector (FPD) x-ray image guidance the catheter was guided into the right coronary artery (RCA). To visualize the anatomy, iodine contrast was manually injected into the artery via the catheter and the injection was visualized under 15 fps cine imaging modality (Figure 1). Following this HSCA was performed.

Figure 1:

Three consecutive images of right coronary artery (RCA) contrast injection from a sequence acquired using conventional FPD 15fps Digital Angiography (Cine) imaging technique. The arrow in A shows the position of RCA. B and C are cropped and zoomed in, to show the RCA anatomy. No flow information can be derived using these images.

2.2. High Speed Coronary Angiography(HSCA)

A computer controlled automatic injector (Legato-110 from KD Scientific) was connected to the proximal end of the guide catheter and used to inject iodine contrast agent at a steady rate of 40ml/min for a duration of 1 s (total injection volume 0.66 ml). Aries (Direct Conversion, Stockholm), a single photon counting direct conversion detector with 100 μm pixel size and capable of acquiring x-ray images at 1000 fps was used to capture the contrast injection. The detector has a sensitivity of 1.2 counts/μR (136 counts/μGy) for each pixel for the exposure parameters used. The detector was mounted on a mechanical changer unit attached to the frontal C-Arm of the Infinix Biplane C-arm system from Canon Medical Systems Corporation (refer to Figure 2). The exposure was kept at 86 kV, 400 mA for a duration of 1s. From the entire image sequence, only the section of the images that showed flow of contrast was used for analysis.

Figure 2:

The HSA detector is mounted on a changer (white arrow) and can be deployed into the active FOV in front of the FPD when needed using a pre-programmed switch.

2.3. Image Noise Reduction

The HSA detector is a PCD and has zero instrumentation noise; any noise in the image is due to variation in the input x-ray quanta. To reduce this noise in the image, a machine learning model using Long Short Term Memory (LSTM) network⁴ and attention mechanism⁵ was built. Figure 3 shows the network architecture. The network is designed to reduce noise in a sequence of 300 images which is typically the period of systole in humans. The network takes in a sequence of 300 time points, with 25 (5×5 neighborhood) pixel values in each time point and generates a sequence of 300 times points with 1 pixel value for each time point. The illustration presented in figure 4 shows how the images are first restructured to the neural network input format. After prediction the output is then restructured back to constitute the image sequence HSA images of iodinated microspheres in a 3D printed phantom from a previously published study³ was used to qualitatively determine if the noise reduction using the model in Figure 3 resulted in any blurring artifacts due to motion. The HSA images were also reduced in noise using standard spatial-temporal filtering methods for comparison. Equation 1 and 2 describe the spatial and recursive filter (weight = 0.1) applied to each pixel in the image (first spatial then recursive).

Figure 3:

LSTM network with attention mechanism. The network is designed to reduce noise in a sequence of 300 images at a time.

Figure 4:

The input image sequence is first restructured according to the input requirements of LSTM network prediction. After prediction of the entire dataset, the predicted output is restructured back to the generate the noise reduced image sequence.

$$I_{{\mathit{\text{out}}}_{\mathit{\text{spatial}}}}(x,y,t)={\sum }_{i=-2}^{i=2}{\sum }_{j=-2}^{j=2}{I}_{\mathit{\text{Noisy}}}(x+i,y+j,t)\times {\mathit{\text{Kernel}}}_{\mathit{\text{Gaussian}}}(i,j)$$ (1)

$$I_{{\mathit{\text{out}}}_{\mathit{\text{recursive}}}}(x,y,t)=(\mathit{\text{Weight}})\times I_{{\mathit{\text{out}}}_{\mathit{\text{spatial}}}}(x,y,t)+(1-\mathit{\text{Weight}})\times I_{{\mathit{\text{out}}}_{\mathit{\text{recursive}}}}(x,y,t-1).$$ (2)

The noise reduced images for this comparison study are presented in Figure 5.

Figure 5:

Comparison of lag/motion blur artifacts post noise reduction in a 1000 fps High Speed Angiogram of iodinated microsphere injection. From a sequence of 300 images (300 1ms timepoints) two images (A and D) 20 ms apart are used. Arrows point to regions of higher differences in lag artifacts between the two noise reduction methods.

A: 1 ms HSA image before processing. Red box shows the ROI location used to calculate signal to noise ratio in all images, B: Spatio-Temporal filter output of image A., C: LSTM Neural network output of image A.

D: 1 ms HSA image 20 ms after image A, E: Spatial-Temporal filter output of image D.,F: LSTM Neural network output of image D

2.4. Quantitative analysis

From the HSCA acquisition of the contrast injection, a 300 image sequence showing flow of contrast was used for analysis. First the images were reduced in noise using the neural network presented in Figure 3. Following this, the velocity distribution of contrast flow through the artery was derived from the noise reduced images using previously published optical flow (OF) methods.⁶ The method calculates the OF between two images, so for a sequence of 300 images the algorithm calculates 299 velocity profile images. From these images a projection image of maximum velocities recorded per pixel for the entire segment was derived to track flow throughout the vessel. The diameter of the guide catheter (5 F = 1.67 mm) was used as a reference to obtain a conversion factor from pixel distances to mm.

A total of 3 HSCA runs were performed in the Right Coronary artery. The noise reduced images of contrast flow and the corresponding maximum velocity profile map for the injection is presented in the results section.

3. RESULTS

The comparison of lag/motion blur due to the two noise reduction methods in the High Speed Angiograms of microsphere particles is presented in Figure 5. It can be seen that the standard spatial-temporal filtering method introduces significant lag and motion blur, while such effects are minimal when the neural network is used. The signal to noise ratio for the ROI shown in Figure 5 was calculated to be 5.36 for the image with noise, 107.19 for the corresponding ROI in the image reduced in noise using spatial-temporal filtering and 181.3 for the image reduced in noise using the neural network presented above.

The result of noise reduction on one of the HSCA images of the swine right coronary artery using neural network is shown in Figure 6. It can be seen that after noise reduction contrast bolus is more easily visible compared to its immediate background. A plot of the pixel values of the ROI in Figure 6A before and after the two noise reduction methods is shown in figure 6. It can be seen that the contrast bolus arrival in the neural network noise reduced images follows very closely to the noisy images, whereas a lag is introduced when the noise is reduced using the spatial-temporal filtering method. The velocities within the coronary artery range between 25–35 cm/s with higher velocities recorded in the vessel center.

Figure 6:

Demonstration of Noise reduction using LSTM network on images acquired during 1000 fps High Speed Coronary angiography (HSCA). Image A and C : 1 ms images 30 ms apart. Image B and D: LSTM Network Noise reduced output images of A and C, respectively. The arrow in B shows the direction of flow. The graph shows a temporal plot of normalized mean values for the ROI show in figure A, before and after the two noise reduction methods. The graph shows that while spatial and temporal filtering does reduce noise, it also introduces lag which can lead to lower velocity calculation. The LSTM network follows the input more closely while reducing noise. This is consistent with the results presented in figure 5.

Figure 7, 8 and 9 show 3 images from a sequence of 300 images, followed by velocity distribution of contrast/blood flow for three separate HSCA runs imaging the right coronary artery in a single swine model. A mean value of 35 counts per pixel was measured in the ROI shown in figure 7B for the HSCA runs shown in figures 7,8 and 9 The detector entrance air kerma (including scatter) for each sequence of 300 images was calculated to be 8.74 mR (77 μGy) at 29 μR/frame (300 images × 35 counts per pixel /1.2 count per μR). This is equivalent to a typical clinical DSA acquisition at 3 fps for 12 seconds, with a typical detector entrance air kerma of 250 μR/image (3 f/s × 12 s × 250 μR/image / 1000 = 9.0 mR). At the measured 25–35 cm/s, 300 frames is sufficient for the contrast bolus to travel over a vessel length of 7.5–10.5 cm and thus for the velocities to be calculated over that length.

Figure 7:

Demonstration of 1000 fps High Speed Coronary Angiogram #1. Iodinated contrast was injected at a steady rate of 40 ml/min for a duration of 1 s. A total of 1000, 1 ms images of the injection were acquired. From these images, a sequence of 300 1 ms images where flow of contrast is higher is manually selected. A-C: 1 ms noise reduced (LSTM Network) images showing flow of contrast bolus (follow arrows) through the right coronary artery. A mean of 35 counts/pixel was measured in the ROI shown in figure B. D: Velocity profile derived from the image sequence of 300 noise reduced images (LSTM Network) using the previously published Horn Schunk OF method. The velocities were calculated over 1500 iterations. The velocities within the coronary artery range between 20–35 cm/s with higher velocities recorded in the vessel center. The region shown by the white arrow shows velocities recorded due to movement of the catheter.

Figure 8:

Demonstration of 1000fps High Speed Coronary Angiogram #2. Iodinated contrast was injected at a steady rate of 40 ml/min for a duration of 1s. A total of 1000, 1 ms images of the injection were acquired. From these images a sequence of 300 1 ms images where flow of contrast is higher is manually selected. A-C: 1 ms noise reduced (LSTM Network) images showing flow of contrast bolus through the right coronary artery. D: Velocity profile derived from the image sequence of 300 noise reduced images (LSTM Network) using the previously published Horn Schunk OF method.

Figure 9:

Demonstration of 1000 fps High Speed Coronary Angiogram #3. Iodinated contrast was injected at a steady rate of 40 ml/min for a duration of 1 s. A total of 1000, 1 ms images of the injection were acquired. From these images a sequence of 300 1 ms images where flow of contrast is higher is manually selected. A-C: 1 ms noise reduced (LSTM Network) images showing flow of contrast bolus through the right coronary artery. D: Velocity profile derived from the image sequence of 300 noise reduced images (LSTM Network) using the previously published Horn Schunk OF method. The velocities were calculated over 1500 iterations. The velocity distribution is similar to other velocity distributions shown in figures 7D and 8D

4. DISCUSSION

Coronary Artery Disease (CAD) is the most common type of heart disease which causes dangerous thickening and narrowing of coronary vessels. This can disrupt the flow of blood carrying oxygen and nutrients to the heart, causing serious life threatening problems. Percutaneous Coronary Intervention (PCI) is a well-established endovascular diagnostic and treatment procedure for CAD. During these procedures, typically a catheter is guided to the proximity of the diseased vascular anatomy under x-ray imaging. This is followed by coronary angiography which involves an iodine contrast injection in the target coronary vessel performed under either digital angiography (DA or Cine, at least 7.5 fps). Due to poor temporal resolution from heart motion, Digital Subtraction Angiography (DSA) imaging is not often used during PCIs.

Coronary angiography remains the default method to define coronary anatomy and characterize the severity of coronary arterial stenoses. Using these images, the percentage of stenosis defined as the ratio of vessel diameter in the diseased area to the diameter of the same vessel in a healthy area is calculated and used as a measure to gauge the severity of the disease. As per the AHA guidelines a visually estimated diameter stenosis severity of ≥ 70% has been used to define significant stenoses warranting revascularization treatment.⁷

The guidelines also state that any angiographically determined stenosis between 40% to 69% warrants additional investigation to assess the physiological significance of the stenosis. For this subset of patients, use of fractional flow reserve (FFR) is recommended to guide the decision to proceed with Percutaneous Coronary Intervention (PCI) for revascularization. High Speed 1000 fps Coronary Angiography (HCSA) presented in this work can be used to derive physiological information such as flow velocity. The HSCA technique can be applied to derive velocity profiles through any artery structure, including regions distal and proximal to stenosis. The change in flow velocity immediately before and after the stenosis can be used as a measure to evaluate the severity of the stenosis and determine the need for PCI for revascularization.

FFR measurement on the other hand is an invasive endovascular procedure, requiring the use of specially designed wires with sensors that are used to measure pressure differential before and after stenosis. In order to do this, the pressure wires have to be maneuvered to and beyond the area of stenosis. The presence of a device in a restricted vessel structure could affect its measurement. Also prior to measurement the patients could have administered drugs such as adenosine to induce hyperemia to ensure there is enough blood flow in the vessel to get good measurements. Unlike FFR, the HSCA presented in this work was able to derive velocity measurements in coronary arteries from routine contrast injection procedures without the use of invasive measurement devices.

5. CONCLUSIONS

This is the first in-vivo live study that attempts to derive coronary flow velocities and velocity distribution profiles within a vessel using diagnostic angiography procedures, without the use of any dedicated interventional devices such as a pressure wire, or an ultrasound wire and at doses comparable to doses for conventional angiographic sequences. Further investigations are warranted to expand our understanding of the applications of HSCA in cardiovascular research and clinical practice.