A Novel Computer Image Analysis Algorithm Improves Inter-Rater Reliability in the Evaluation of Respiratory Mucosal Ciliary Ultrastructure.
Keith Funkhouser, Marc Neithammer, John Carson, Brock Baker, Susan Minnix, Margaret Leigh, Michael Knowles, William Funkhouser. UNC, Chapel Hill, NC
Background: Primary ciliary dyskinesia (PCD; MIM 242650) is a genetic disease causing defects in the structure or function of cilia. It is characterized by recurrent infections of the upper and lower respiratory tract due to reduced mucociliary clearance. Presently, diagnosis requires identification of dynein arm loss in transmission electron microscopy (TEM) images. These images exhibit highly variable background signal in the axoneme, which can confound interpretation of dynein arm presence or absence. We developed a novel computer image analysis algorithm in the MATLAB(c) environment to align axonemal microtubular pairs. The purpose of this study was to determine whether inter-rater reliability of PCD diagnosis could be improved through the use of this method.
Design: In a randomized, double-blind experiment, two morphologists evaluated dynein arm loss in 32 cases of possible PCD. Each evaluated the presence of inner and outer dynein arms using three distinct methods. In the first method, each morphologist was given all of the scanned TEM images for the cases. For the second and third methods, 10 in-focus axonemes from each case were processed by our software. In the second method, each morphologist was given 10 averaged images (1 averaged image of the 9 microtubular pairs for each of 10 axonemes) for each case. In the third method, each morphologist was given a single averaged image of all 90 microtubular pairs in the 10 axonemes for each case.
Results: Cohen's kappa (Κ) was calculated for each of the three methods. For the "gold-standard" TEM technique, we found Κ=0.68. Using 10 computer-generated axonemal averages per case, we found Κ=0.81. Using only the single computer-generated average of all microtubular pairs per case, we found Κ=0.94. An example of a computer-generated average displaying inner dynein arm loss is shown below.
Conclusions: Our novel image alignment algorithm improves inter-rater reliability for evaluation of dynein arm loss in ciliary axonemes. Studies are underway to determine whether the accuracy of TEM-based PCD diagnosis is also improved.
Wednesday, March 2, 2011 9:30 AM
Poster Session V # 267, Wednesday Morning