In Vivo Measures Of Key Parameters For Cochlear Implantation

Cochlear implantation (CI) has become a standard clinical intervention performed worldwide for profound sensorineural hearing loss. Many studies have shown that the speech recognition of implant users can be improved greatly when the frequency band assigned to each electrode is matched onto the cochlear location that normally processes that frequency range. It has been found that the human cochlea's frequency response characteristic is exponentially distributed along the basilar membrane. Based on this fa t, to estimate the acoustic frequencies to which nearby auditory neurons are most sensitive, the total length of a patient's cochlea and the insertion depth of each electrode must be accurately measured. It is challenging to accurately measure the cochlea length in clinical applications for the following difficulties: ?the cochlea is a very small and complex anatomical structure, ?segmenting the cochlea from the medical images is very difficult, ?lacking a method for accurately measuring the cochlea length. Additionally, the pre-operative cochlea implant surgical planning is crucial for the success and security of the surgery. Because the resolution of the medical images is low, the inner ear is very small and complex anatomical structure, and so on, therefore it is difficult for doctors to form a correct diagnosis on the malformation of the inner ear only by viewing the multi-spiral computed tomography (MSCT) images slice by slice and it is even more difficult to ascertain whether the scala tympani and the scala vestibuli of the cochlea are unblocked throughout their length. Therefore, how to obtain the pre-operative and post-operative key information for the CI using the MSCT images of the temporal bone has become a problem urgently needing to be solved in the clinical applications. The key techniques studied in this thesis include the cochlea segmentation, the cochlea length measure, the orthogonal cross sections extraction, the inner ear segmentation, the visualization of the inner ear, the insertion depth measure of the electrode array, and so on. The main achievements of this thesis are as follows.1. For cochlea segmentation, [1] an interactive, semiautomatic, coarse-to-fine segmentation approach with volume rendering feedback to separate the cochlea is presented in our thesis. [2] The 3D narrow-band level set algorithm based on region competition is modified and is applied for the fine segmentation of the cochlea, which facilitates the adjustment of the intermediate parameters and reduces the segmentation time. [3] The correlation of the segmented result to the image volume is built to evaluate the validity of the segmented cochlea. The accuracy can meet the requirements of clinical application. [4] The cochlea segmentation is automatically finished except for introducing user interaction to locate the initial contour and adjust the parameters; [5] The total time of our segmentation method takes from 3 to 5 minutes to obtain a cochlea model,whereas that of using the 2D deformable active contour model algorithm to perform the sectional images slice by slice takes from 15 to 25 minutes, that of manual delineation takes even several hours. Moreover, the cochlea separated by our method is of smooth surface.2. For the cochlea length measure, [1] a method combining the automatic tracking algorithm with manual specification to measure cochlea length is provided in our thesis. [2] A new efficient automatic tracking algorithm based on distance transform is designed to extract the centerline of the basal turn. [3] The middle and apical turns are difficult to be separated from each other. The orthogonal cross sections along the cochlea length are extracted in turn, and the mass center of the cochlea is specified by the user interaction. The correlation of the cochlea canal to its cross section is built to improve the accuracy of the manually specified centroids. [4] At last, a Cardinal spline function is used to approximate all center points to get a smooth centerline. The total length of the cochlear canal is calculated by summing up the incremental arc lengths along the central path. [5] Compared with the parameterized helico-spiral model approximation method, the cochlea length measured by our method is of higher accuracy. This conclusion is proved intuitively by using the correlation of the extracted centerline to the cochlea canal in 3D. The accuracy can meet the requirements of clinical application. Moreover, our method is faster than the manual delineation method, and it can be used not only for normal cochlen but also for cochlea with malformation.3. For the inner ear segmentation, [1] the modified 3D narrow-bma levei set algorithm based on region competition is used for the segmentation of the inner ear. [2] The inner ear is separated quickly by locating he approximate initial contour to the target in 3D and adjusting the intermediate parameters. Five inner ears are segmented using the clinical temporal bone MSCT images, four of them are segmented automatically, one result, which is Mondini malformation, contains more than the objects of interest. [3] The correlation of the point on the resultant surface to the three orthogonal sections that intersect at that point on the surface is built to find the weak edges quickly. The unwanted structures are deleted manually on the sections with weak edges. [4] Our segmentation method takes about 10 minutes to obtain a complete inner ear model, whereas that of manual delineation or automatic algorithm followed by manual modification takes several hours. Moreover, the inner ear extracted by our method is of even surface.4. For the electrode array insertion depth measure, the centerline of the electrode array is extracted by applying the automatic centerline extraction algorithm designed in our thesis. The length of the central path is the insertion depth. This insertion depth measure method eliminates the negative influence introduced by partial volume effects, except the entry and end points. The MSCT images of one implant recipient are used to objectively measure the insertion depth of the electrode. The objective and accurate result measured by our method is 16.14mm, whereassurgical observation is 16~17mm. With the measured cochlea length and electrode array insertion depth, the characteristic frequencies of the most apical electrode position and the entry point are estimated by using cochlea frequency-place function. The estimated frequencies are valuable for frequency band assignments in patient's speech processor program.5. In vivo studies about the scala vestibuli and scala tympani based on MSCT images are carried out in this thesis. The orthogonal cross sections along the cochlea are extracted with the centerline of the cochlea (or electrode array). Cross sections offer the opportunity for diagnosis and preoperative assessment of patients's scala tympani and scala vestibuli, which is important for choosing the best channel into which the electrode array will be inserted. With the post-operative cross sections, doctors can ascertain whether there is pyogenic infection, determine the distribution of the electrode in the cochlea, which is helpful to improve the surgery process and the design of the electrode array.