Automated analysis of vasculature in coronary angiograms
A system analyzes angiogram image data, including video data, in order to extract vasculature information. Advanced image processing and machine learning techniques are used in pre-processing, frame selection, and vasculature segmentation to remove classes of artifacts from angiogram videos, and specifically from frames selected as having a sufficient amount of image data. From segmentation, accurate vasculature diameters are calculated, and, in some examples, stenoses and/or the extent of stenosis is automatically determined and displayed.
1. A computer-implemented method for analyzing medical video image data for a subject, the video image data being formed a set of frames of medical image data, the method comprising:
obtaining, at one or more processors, the medical video image data and performing, at the one or more processors, a pre-processing on the obtained medical video image data by performing on each frame of the video image data
(i) a denoise filtering on the obtained medical video image data,
(ii) a removal of a first set of features, and
(iii) an image quality assessment indicating an amount of usable image data that appears in the frame;
automatically selecting, at the one or more processors, a subset of the frames of medical image data by performing on each frame of the video image data,
(i) a vessel segmentation,
(ii) a histogram analysis after the vessel segmentation, and
(iii) determining an amount of visible vasculature for each frame, and
further identifying, at the one or more processors, based on the amount of visible vasculature for each frame, frames having a desired amount of visible vasculature as the subset of frames of medical image data;
automatically removing, at the one or more processors, artifacts from the subset of frames of medical image data using a shape characteristic machine learning engine trained using a set of artifact training data, wherein the shape characteristic machine learning engine applies a width profile analysis on identified features in the subset of frames to determine if any of the identified features are artifacts, in which case the artifacts are removed;
automatically performing, at the one or more processors, segmentation on the subset of frames with artifacts removed, wherein the segmentation comprises an extraction of the vasculature of each of the subset of frames;
automatically performing, at the one or more processors, a width profile analysis on one or more portions of the vasculature in each of the subset of frames to determine an amount of stenosis in the vasculature; and
aggregating, at the one or more processors, the width profile analyses for the subset of frames to determine of an overall stenosis for the subject.
2. The method of claim 1 , wherein performing a pre-processing on the obtained medical video image data comprises performing contrast adjustment on each frame to increase contrast.
3. The method of claim 1 , wherein performing a pre-processing on the obtained medical video image data comprises performing the denoise filtering by (i) blocking each pixel and (ii) applying a Gaussian smoothing filter is convolved with each block to reduce the noise.
4. The method of claim 1 , wherein performing a pre-processing on the obtained medical video image data comprises performing the denoise filtering by applying a non-local means (NLM) filtering by replacing, for each frame, each pixel of each frame a weighted average pixel intensity of a set of closest nearest pixels, wherein the weighted average pixel intensity is determined from a quadratic pixel distance between the pixel and the set of closest nearest pixels.
5. The method of claim 1 , wherein performing a pre-processing on the obtained medical video image data comprises performing a bilinear interpolation on the pixels of each frame.