IMPROVED CLASSIFIER IN DETECTION OF HEALTHY TENDON REGIONS BY LOOKING AT PARAMETERS IN THE SPATIAL FREQUENCY DOMAIN

Theo Joseph Author

08/05/2020 Added

13 Plays

Description

Video Poster Presentation

Searchable Transcript

Search:

[00:00:01.478]Hey everyone, My name is Kunjan Theodore
[00:00:04.198]Joseph and I am a junior in Biomedical
[00:00:06.880]Engineering at the University of Nebraska
[00:00:09.539]-Lincoln. The topic of my UCARE project
[00:00:12.746]is IMPROVED CLASSIFIER IN DETECTION OF
[00:00:15.676]HEALTHY TENDON REGIONS BY LOOKING AT
[00:00:18.643]PARAMETERS IN THE SPATIAL FREQUENCY
[00:00:21.194]DOMAIN. I work at the Biomedical
[00:00:23.459]Imaging and Biosignal Analysis Lab
[00:00:28.365]under my mentor Benjamin Hage and advisor
[00:00:31.215]Dr Greg Bashford. In order to begin I'll
[00:00:35.625]start with some background about tendons
[00:00:38.420]and tendinopathy. Tendinopathy is a common
[00:00:42.548]musculoskeletal injury and is
[00:00:44.418]characterized by pain in the tendon
[00:00:46.838]and reduced functionality.
[00:00:48.629]Tendons are regions where the muscle
[00:00:51.151]joins the bone and are designed to
[00:00:53.387]handle loading. Currently, very poor
[00:00:55.984]correlations exist between tendon
[00:00:59.381]structure and tendon health due to the
[00:01:01.988]lack of computational methods.
[00:01:04.789]The primary structure of the tendon
[00:01:07.038]consists of parallel arrangement
[00:01:09.188]of the collagen bundles.
[00:01:11.998]A healthy tendon region, viewed in
[00:01:15.219]ultrasound, has characteristic parallel
[00:01:17.629]striations which make it easily
[00:01:19.909]distinguishable from surrounding tissue.
[00:01:23.403]When a tendon suffers damage which
[00:01:25.914]may be due to excessive or sudden loading,
[00:01:28.544]it results in misalignment of the collagen
[00:01:31.449]bundles. By computationally measuring
[00:01:35.109]the degree of misalignment, we can make
[00:01:37.625]comparisons between normal tendons
[00:01:40.027]and those with symptoms of tendinopathy
[00:01:42.497]which can be used to classify tendon
[00:01:45.177]health and can lead to forming a grading
[00:01:47.943]scale for tendon for assessing tendon
[00:01:50.233]health. For this quantification,
[00:01:52.440]spatial frequency analysis is used
[00:01:54.820]which has given rise to spatial frequency
[00:01:57.490]parameters. The current project focuses
[00:02:00.531]on improving the accuracy of one such
[00:02:02.955]parameter called “P6 Width” which is
[00:02:06.751]the average of vertical and horizontal
[00:02:09.480]widths of a 2D Gaussian mesh fitted
[00:02:12.645]to the spatial frequency peak of
[00:02:14.905]the 2D Fast Fourier Transform
[00:02:19.189]Improving the accuracy of P6
[00:02:21.319]Width would allow better identification
[00:02:24.140]and distinction of tendon regions
[00:02:26.499]and arrangement.
[00:02:29.239]Moving on to the methods of data
[00:02:31.399]collection and analysis
[00:02:33.058]The above mentioned spatial frequency
[00:02:35.720]parameter was calculated by a MATLAB®
[00:02:38.288]program developed by BIBA lab.
[00:02:41.829]Data was collected from multiple
[00:02:44.169]regions of tendon Ultrasound images
[00:02:46.459]and characterized as Good
[00:02:48.192]and Bad tendon regions
[00:02:50.008]as well as non-tendinous regions.
[00:02:52.458]We tested several methods of determining
[00:02:55.280]P6 Width such as Elliptical Gaussian
[00:02:57.646]fit and Rotated Elliptical Gaussian fit
[00:03:01.502]Response variables were derived from
[00:03:03.802]Gaussian fits of the data and includ Ustd
[00:03:06.324], Vstd and RMS. The equation here
[00:03:10.324]shows how RMS is derived from
[00:03:12.954]ustd and vstd.
[00:03:15.681]These are a measure of how
[00:03:17.541]accurately a Gaussian fit is generated
[00:03:19.841]from which mathematical relations can be
[00:03:22.671]established. A flowchart of the method
[00:03:28.169]of calculating P6 Width is shown in Figure
[00:03:31.159]1. Figure 1 A shows an ultrasound image
[00:03:39.349]of the Achilles tendon. The green
[00:03:41.820]polygon encloses the tendon region
[00:03:44.866]whereas the rest is the non-tendinous
[00:03:47.373]region. The yellow square box denoting
[00:03:49.898]the region of interest or ROI
[00:03:52.635]contains an example of a good tendon
[00:03:55.508]region
[00:03:58.509]Next figure 1 B is zooms into the ROI
[00:04:01.912]showing the characteristic echo-texture
[00:04:04.782]of the connective tissue fibrils. Notice
[00:04:08.292]the 4 slightly tilted parallel lines
[00:04:13.690]Figure 1 C shows a 2D Fast Fourier
[00:04:19.921]Transform spectrum of the zoomed ROI
[00:04:22.601]For those who do not know what Fourier
[00:04:25.119]Transform is, it is a method to analyze
[00:04:29.119]the frequencies present in an image.
[00:04:32.193]Finally, we come to Figure 1 D,
[00:04:35.425]which shows a 2D Gaussian mesh fitted
[00:04:38.785]to one of the spatial frequency peaks from
[00:04:41.468]from figure 1 C.
[00:04:44.328]Next moving on to the results,
[00:04:48.628]Each of the graph compares the elliptical
[00:04:51.444]Gaussian fitting method shown in blue
[00:04:56.084]and the rotated Elliptical Gaussian
[00:04:58.159]fitting shown in orange.
[00:05:00.074]Figure 2 shows the graphs for Ustd
[00:05:04.804]which is the standard deviation of the
[00:05:07.234]Gaussian mesh fits in the horizontal or
[00:05:09.624]u- direction. For good bad and
[00:05:14.004]non-tendinous regions
[00:05:16.621]Figure 3 shows the standard deviation
[00:05:19.335]of the Gaussian mesh fits in the vertical
[00:05:22.890]or v- direction, for each of the
[00:05:25.094]good bad and non-tendinous regions
[00:05:29.191]Figure 4 shows the graphs for the root
[00:05:33.138]mean square or RMS values of the
[00:05:35.710]Gaussian mesh fits for the
[00:05:39.710]good bad and non-tendinous regions
[00:05:43.785]The asterisk sign represents that the
[00:05:47.075]results are significantly different with
[00:05:49.410]a confidence level of 95%. For the RMS,
[00:05:53.437]the rotated elliptical gaussian had
[00:05:56.511]significantly lower RMS than the
[00:06:00.821]non-rotated elliptical gaussian
[00:06:03.296]which means it is a better fit
[00:06:06.836]Now lets see what these results mean,
[00:06:11.296]Previous work suggests that a lower RMS
[00:06:14.434]correlates to a narrower spatial frequency
[00:06:17.154]peak which correlates to smoother and
[00:06:19.844]healthier connective tissue fibril i.e.
[00:06:23.383]a healthier tendon. So
[00:06:25.743]in comparison, the Rotated
[00:06:27.833]Elliptical Gaussian fitting showed
[00:06:31.225]better distinguishing capability
[00:06:34.564]for different tendon regions than the
[00:06:36.937]elliptical gaussian fitting.
[00:06:40.583]Future work on this project would be
[00:06:42.863]to test the new P6 method for reliability
[00:06:46.263]and repeatability and then incorporate it
[00:06:49.943]in the software with other spatial
[00:06:52.066]frequency parameters to investigate
[00:06:55.073]further relations. Maximum amplitude
[00:06:57.706]percentage is one such parameter which
[00:07:00.156]has shown prospect to be combined with P6
[00:07:04.816]width, in order to help to classify tendon
[00:07:07.396]regions better. In Figure 5. Graph
[00:07:14.696]shows the range of maximum amplitude
[00:07:17.086]percentage of spatial frequency peaks for
[00:07:19.536]"good" tendon regions in blue
[00:07:21.546]"bad" tendon regions in orange
[00:07:23.816]and non-tendinous regions in gray.
[00:07:30.306]There was a significant difference in
[00:07:35.044]results for each pair of regions
[00:07:36.944]with a confidence level of 95%
[00:07:38.943]Now Bringing this presentation to an end
[00:07:41.423]with some worthy acknowledgments
[00:07:43.643]I would like to thank BIBA Lab members
[00:07:46.073]for their guidance and motivation and
[00:07:48.450]would also like to acknowledge the
[00:07:50.434]support of UCARE in funding this
[00:07:52.342]research. And thank you all for listening.

The screen size you are trying to search captions on is too small!

You can always jump over to MediaHub and check it out there.

Comments

0 Comments

IMPROVED CLASSIFIER IN DETECTION OF HEALTHY TENDON REGIONS BY LOOKING AT PARAMETERS IN THE SPATIAL FREQUENCY DOMAIN

Description

Searchable Transcript

Comments icon comment

Related Channels

Comments