Post on 20-Apr-2018
transcript
Digital Image Processing for Diagnostics of Middle Ear Inflammation Jessica L’Heureux, Georgia McAdams, Joanna Wycech
Faculty Advisor: Professor Lin Cheng
Trinity College, 300 Summit Street, Hartford, CT, 06106, USA
Results
Training set: 23 images
Testing set: 10 images
9/10 images were correctly diagnosed
90% accuracy rate
Abstract Otitis Media, an ear infection present in children, is one of the most
commonly misdiagnosed infections in the country; a large
contributing factor to this is the use of otoscopes with the naked eye
for diagnosis. Recently, digital otoscopes have been created that
give the doctor the option of taking digital images that can then be
examined more closely and in greater detail. The objective of this
senior design project was to develop a program in MATLAB using
already existing image processing kernels and techniques, which
would determine if an infection is present and whether it is Acute
Otitis Media or Otitis Media with Effusion. This will provide a reliable
second opinion for a physician to make his diagnosis. Images taken
by Dr. Tulio Valdez from Hartford Children’s Hospital with digital
video otoscopy devices were used in this project. Filtering
techniques, segmentation, light levels, analyzing several biological
landmarks that are characteristic of this infection were all used to
create a final product that can detect Otitis Media with reasonable
accuracy.
REFERENCES & ACKNOWLEDGMENTS Professor Lin Cheng, Trinity College
Professor John Mertens, Trinity College
Dr. Tulio Valdez, Hartford Children’s Hospital
Alberti, P.W. The Anatomy and Physiology of the Ear and Hearing. Retrieved from
http://www.who.int/occupational_health/publications/noise2.pdf
Thrasher, R.D. (2013, April 25). Otitis Media With Effusion. Retrieved from:
http://emedicine.medscape.com/article/858990-overview#showall
Rachamanidou, A. Otitis Media. Retrieved from: http://www.ent-
matters.co.uk/files/Otitis_media.jpg
Berman, S. Normal Tympanic Membrane. Retrieved from:
http://meded.ucsd.edu/clinicalmed/EarTM.jpg
Keeley M.G. (2011, September). Acute Otitis Media: 6 Steps to Improve Diagnostic Accuracy.
Retrieved from: http://www.pediatricsconsultant360.com/content/acute-otitis-media-6-steps-
improve-diagnostic-accuracy
Rothschild M. Otitis Media with Effusion. Retrieved from:
http://www.kidsent.com/images/clinical/nl_ome.jpg
Cheng, L., Liu, J., Roehm, C.E., & Valdez, T.A. Ear Image Analysis from a Digital Video
Otoscope Prototype using Laser.
McAndrew, A. (2004). Introduction to Digital Image Processing with MATLAB. Thomson Course
Technology.
Bubble Analysis
The presence of bubbles is the main characteristic of OME
This part of the program uses double thresholding to reveal
all possible bubble-like shapes
They are then segmented out and filtered by pixel diameter
and area to determine the final number of bubbles present
This is an accurate indicator of OME
K-Means Clustering
The ability to segment out the auditory bone often rules out the
presence of an infection
K-Means color based segmentation is used to determine if the
bone is visible
The image is separated into specified number of clusters and
grouped based on the Euclidean distance between each point
and the centroid of the cluster
Cluster 3 is further filtered and evaluated to determine if the ear is
healthy or unhealthy
Background
The middle ear consists of the tympanic membrane, cavity, and
bones
Otitis Media is the inflammation of the middle ear and can be
separated into two categories:
Otitis Media with Effusion (OME) – the ear becomes filled
with a thick fluid
Acute Otitis Media (AOM) – the tympanic membrane
become inflamed and bulges outward
Diagnostic Uncertainties include:
Difficult to distinguish between OME & AOM
AOM is frequently over diagnosed
Central Concavity
The presence of a donut-like shape is the main characteristic
of AOM
After filtering, this part of the program allows a physician to
manually zoom in on the circular abnormality and place a
cursor in the center
Then the pixel intensity is evaluated at that specified point
If the intensity is less than the threshold, this indicates AOM
is present
If the intensity is greater than the threshold, this indicates no
central concavity
Figure 1: (a) Healthy Middle Ear (b) OME (c) AOM
Figure 4: Bubble Analysis
Figure 2:Color Based Segmentation K-Means Clustering of a Healthy Ear
Figure 3:Cluster 3 Evaluation for (a) Healthy Ear (b)OME (c) AOM
Figure 4: (a) Central Concavity Present (b) No Central Concavity Present
Vascularity
Branching vascularity is a prominent feature in OME,
whereas in AOM parallel alignment of vessels is present
along the donut
Matched filter detection is used to detect piecewise linear
segments of blood vessels
The green channel in the RGB image is selected for
processing, and after the filter kernel is applied only the
maximum of the response is retained
Based on the range of the response, categorization of the
infection type is made
Figure 5: Vessel Extraction in a AOM Infected Ear
Truth Table for Results
The truth table for results represents the process by which
a healthy ear, OME, or AOM was detected through the
MATLAB program
Table 1: Truth Table for Results