Automatic segmentation and tracking of biological prosthetic heart valves

Abstract

Purpose: Prosthetic heart valve designs must be rigorously tested using cardiovascular equipment. The valve orifice area over time constitutes a key quality metric which is typically assessed manually, thus a tedious and error-prone task. From a computer vision viewpoint, a major unsolved issue lies in the orifice being partly occluded by the leaflets' inner side or inaccurately depicted due to its transparency. Here, we address this issue, which allows us to focus on the accurate and automatic computation of valve orifice areas. Approach: We propose a segmentation approach based on the detection of the leaflets' free edges. Using video frames recorded with a high-speed digital camera during in vitro simulations, an initial estimation of the orifice area is first obtained via active contouring and thresholding and then refined to capture the leaflet free edges via a curve transformation mechanism. Results: Experiments on video data from pulsatile flow testing demonstrate the effectiveness of our approach: a root-mean-square error (RMSE) on the temporal extracted orifice areas between 0.8% and 1.2%, an average Jaccard similarity coefficient between 0.933 and 0.956, and an average Hausdorff distance between 7.2 and 11.9 pixels. Conclusions: Our approach significantly outperformed a state-of-the-art algorithm in terms of evaluation metrics related to valve design (RMSE) and computer vision (accuracy of the orifice shape). It can also cope with lower quality videos and is better at processing frames showing an almost closed valve, a crucial quality for assessing valve design malfunctions related to their improper closing.

Keywords: active contours; leaflet edge detection; motion pattern; orifice area segmentation; prosthetic heart valves; video analysis.

Fig. 1

Example frames from a test video, showing the opening motion of a tricuspid prosthetic valve: (a) valve closed; (b) partly open; and (c) completely open.

Fig. 2

Example frames in which the valve orifice area is partially visually bordered by the inner side of the leaflets (white arrows) instead of the free edges of the leaflets.

Fig. 3

Flow chart of the proposed method.

Fig. 4

Pre-processing: (a) original image; (b) circular outline of the valve (ROI); (c) various potential triangular masks; (d) triangle with lowest average gray value; and (e) final triangular initial mask $M_0$ for active contouring.

Fig. 5

Combining active contouring and thresholding. (a) and (c) Segmented orifice area pixels for two different cases, with white indicating pixels segmented by both methods, magenta indicating pixels segmented by active contouring only, and green indicating pixels segmented by thresholding only. (b) and (d) The corresponding borders of the combined segmented pixels are shown in red, superimposed over the original frames.

Fig. 6

The start of the opening phase of the valve is determined from a sudden increase (Region 1) followed by a sudden drop (Region 2) of the average gray-level value of the frame.

Fig. 7

Triggering active contouring: average gray-level value inside the maximal circle fitted within the initial triangular mask, at different frames of the opening phase. The last circle on the right corresponds to Frame A.

Fig. 8

Leaflet free edge detection: (a) cropped region of initial frame, (b) examples of short fitted curves along boundary pixels (green), corresponding orthonormal curves (blue), and segmented orifice area (black), (c) full orthonormal curves (blue), (d) modified orifice area along orthonormal curves (white) when considering only the absolute maximal brightness, and (e) detected free edge pixels (red) when considering also the local maximal brightness and smoothness of the contour. White arrows in (e) point to key refinement locations; the actual free edges were successfully recovered.

Fig. 9

Refining the orifice area along (a) normal lines versus (b) orthonormal curves. The regions inside the red ellipses show less candidate pixels for contour refinement when using orthonormal curves in the case of bright commission points in the vicinity of the boundary pixels.

Fig. 10

Sample GT three-level images. (a) and (c) Cropped region of original frames and (b) and (d) corresponding GT data. Gray = central orifice areas, white = inner leaflet areas, black = other.

Fig. 11

Comparison of the extracted orifice areas over time for OAS, OLFED, and AOCFED (proposed) with the GT data.

Fig. 12

Typical results at different moments in the cycle for OAS, OLFED, and AOCFED (proposed), along with the GT, for PHV A. Frames are cropped for visualization purposes. White arrows indicate issues that our proposed method is able to successfully address.

Fig. 13

Typical results at different moments in the cycle for OAS, OLFED, and AOCFED (proposed), along with the GT, for PHV B. Frames are cropped for visualization purposes. White arrows indicate issues that our proposed method is able to successfully address.

Fig. 14

Impact of frame rate on image bluriness. (a) and (c) Sample (cropped) frames at 250 fps and (b) and (d) corresponding frames at 1000 fps.

Fig. 15

Comparison of the extracted orifice areas over time by our proposed approach (AOCFED) with the GT data, for three different frame rates.