Fast and accurate detection of extraocular muscle borders using mathematical morphology

Extraocular muscle (EOM) measurements are usually performed on 2D CT or MR images. Typically this measurement requires a subjective, inaccurate and time-consuming identification of the EOM. A new algorithm is presented which is based entirely on mathematical morphology, that quantifies and isolates the EOM. Knowing that the EOM has higher intensities than surrounding structures, morphological filters are used to highlight the muscles by smoothing out noises in the orbit without blurring the edges. Only three parameters are set in the whole process. No manual extraction is used. The algorithm successfully extracts EOM borders in no longer than 12 seconds. The same procedure takes about 1 minute or more by manual segmentation. Pearson product-moment correlation coefficients were calculated to examine the relationship between automatic and manual segmentation for area (R=0.92). We have presented a new algorithm to identify and measure small structures like EOM in orbital CT images. It is robust to detect the muscle, accurate and faster than manual segmentation.