Automated pneumothorax detection
A method of determining the presence of a pneumothorax includes obtaining a series of frames of image data relating to a region of interest including a pleural interface of a lung. The image data includes at least a first frame and a second frame. The method further includes identifying, via processing circuitry, the pleural interface in at least the first frame and the second frame, determining, based on computing optical flow between the first and second frames, a pleural sliding classification of the image data at the pleural interface, and determining whether a pneumothorax is present in the pleural interface based on the pleural sliding classification.
1. A method comprising:
obtaining a series of frames of image data relating to a region of interest including a pleural interface of a lung, the image data including at least a first frame and a second frame;
identifying, via processing circuitry, the pleural interface in at least the first frame and the second frame;
determining, based on computing optical flow between the first and second frames, a pleural sliding classification of the image data at the pleural interface; and
determining whether a pneumothorax is present in the pleural interface based on the pleural sliding classification.
2. The method of claim 1 , wherein the determining the pleural sliding classification comprises determining the pleural sliding classification based on clustering of feature vectors indicative of the optical flow based on Euclidean distance to generate clusters that are uniform in direction and spatially adjacent.
3. The method of claim 2 , wherein the determining the pleural sliding classification further comprises employing angular histograms that overlap with the pleural interface of the region of interest using clusters generated by the clustering of the feature vectors.
4. The method of claim 3 , wherein the determining the pleural sliding classification further comprises classifying the angular histograms via a trained support vector machine.
5. The method of claim 4 , wherein the classifying the angular histograms comprises classifying the angular histograms as one of left-sliding, right-sliding and no-sliding.
6. The method of claim 1 , wherein the identifying the pleural interface comprises combining a Hough transform that finds lines in the first and second frames of the image data with brightness thresholding to identify bright lines in the first and second frames of the image data.
7. The method of claim 6 , further comprising employing a spatial constraint giving preference to bright lines closer to the lung to locate the pleural interface.
8. The method of claim 1 , wherein the image sensor is configured to obtain one of one-dimensional, two-dimensional or three-dimensional video imagery of the lung as the image data.
9. The method of claim 1 wherein the determining whether the pneumothorax is present in the pleural interface comprises employing a hidden Markov model to analyze the pleural sliding classification relative to a predetermined set of state transition probabilities.