An overview of the proposed process flow is shown in Fig. 1. Contrast-enhanced X-ray pulmonary angiography images were acquired. The acquired images are incorporated in the image processing flow. The baseline mask image, obtained before contrast agent injection, is subtracted from subsequent, consecutive images. The ROI was determined in each right and left lung region, as shown in Fig. 2. The TIC of two ROIs are obtained. The temporal time window for the analysis of the obtained TIC is optimized by the new algorithm. The parameters of the TIC are calculated within each region for the optimized time window. Finally, the right-to-left ratio is calculated.
Image analysis process flow for right-to-left ratio of blood flow distribution.
Regions of interest to measure lung right and left blood flow.
To achieve objective, quantitative, and reproducible automated methods, several key features are needed. These include accuracy, reproducibility, broad application to different types of diseases and quick computation that is clinically acceptable for the procedure. To address these requirements, in this work, we developed algorithm for determining ROI size, ROI location, automated optimization of the temporal time window, stable selection of parameters in the TIC, and minimization of computational time.
ROI determination
For quantitative assessment of the right-to-left ratio of pulmonary blood flow distribution, rectangular ROIs are placed in the right and left regions, as demonstrated in Fig. 2. When an image is acquired, the field of view and source imager distance are usually adjusted so that the whole lung is maximally included in the image to minimize the patient radiation dose. Therefore, the ROI can be enlarged to cover the left and right lungs; these ROIs are close to the vicinity of the image edges, as shown in Fig. 2. The gap between the right ROI and left ROI at the middle of the image is increased as much as possible so that the main pulmonary trunk and tip of the catheter are excluded, but the whole lung region is included. A larger gap between the left and right ROIs is also beneficial when assessing many complex pediatric treatments, such as Blalock-Taussig shunts. In this study, ROI size and location are fixed for all cases analyzed. In 1024 by 1024 images, the ROI width is 350, and the ROI height is 820. The coordinates are shown in Fig. 2; right ROI (X1, Y1, X2, Y2) = (9, 103, 358, 922) and left ROI (X3, Y1, X4, Y2) = (665, 103, 1014, 922). ROI selection is not impacted by dynamic acquisition because diaphragm motion is not critical during a short period of time within 200 ms. The X-ray image acquisition angle of cranial (CRA) and caudal (CAU) directions can be applied as well as anterior–posterior (AP) directions. However, left anterior oblique (LAO) and/or right anterior oblique (RAO) directions cannot be used.
TIC measurement
Contrast-enhanced XA images are incorporated in image processing. The baseline mask image, obtained before contrast agent injection, is subtracted from subsequent, consecutive images. The TIC of two ROIs were obtained by averaging all pixel values in each ROI. Using this averaging approach to calculate TIC, computational time is dramatically reduced. The original calculation requires image-based processing of all pixels in the image, which corresponds to image width by image height (for example, 1024 by 1024 pixels). However, the current ROI-based processing approach requires only two calculations (left and right ROI).
Time window optimization
The right-to-left ratio of pulmonary blood flow distribution is calculated only in the specific temporal time window to measure equivalent blood flow with LS that has different tracer kinetic models3,9,10. In X-ray angiography, the temporal time window is required to be set at the torrent period during the second cardiac cycle after contrast injection. The torrent period is a short period during which the contrast agent is torrentially discharged from the pulmonary arteries to the capillary bed. The second cardiac cycle is used to eliminate variance in contrast agent concentration because contrast agent is not well mixed and unilaterally distributed in the pulmonary trunk in the first cardiac cycle immediately after contrast injection. This unilateral distribution leads to one side flow in the first cardiac cycle. Using the second cardiac cycle, this variance is reduced, and stable measurement is achieved.
In this paper, the mean TIC combining both the right and left regions is used, and the time of maximum slope of the combined TIC is detected. If one side flow occurred due to unilateral distribution in the pulmonary trunk, the combined TIC would have a small slope because the total amount of contrast flow was small; hence, the time of one side flow would not be detected. If the contrast is well mixed, the contrast agent flows to both the right and left regions simultaneously, the total amount of contrast flow is large, and the combined time-signal density curve should have a steep slope.
The length of the time window is set to less than 200 ms; six frames in the case of 30 frames/s data acquisition. This is because the period from the time when contrast agent arrives at the first branch of the pulmonary artery to the time when contrast agent fills the entire lung field is approximately 200 ms.
We observed that the starting time of contrast flow from the pulmonary trunk was slightly different between the right and left sides. The difference is up to 100 ms. This difference does not affect LS measurements that count temporally accumulated tracer11,12. On the other hand, it impacts the proposed method because the proposed method does not measure accumulation but measures the net increase in TIC in a short time window. In this paper, a new automated algorithm is proposed to achieve stable results even in cases when the contrast flow starting time is slightly different. In this algorithm, the time window is optimized for the right and left lung regions independently. First, a representative six-frame time window is detected by using the above combined TIC. Second, it is extended by eight frames: four frames before and after the representative six frames. A total of 14 frame lengths are determined as a candidate time window. Third, in this 14 candidate frame time window, six frames that show the maximum slope of the TIC are selected in each right and left region independently. These steps are shown in Fig. 3. In summary, optimized time windows are selected for each right and left region independently in the same cardiac cycle.
New algorithm to optimize time window to measure pulmonary blood flow distribution.
Parameter calculation
The right-to-left ratio of pulmonary blood flow distribution is calculated by the net increase in signal intensity and is an equivalent model with scintigraphy3. In this paper, a stable selection of parameters is investigated. If only two points are used to measure net increase in signal intensity, it is easily affected by several noise factors, such as body motion, heart motion, and image acquisition noise. Therefore, in this paper, all six points in the time window are used to calculate the slope using linear fitting. This approach is equivalent to the scintigraphy method, and it makes the algorithm stable and robust.
