Tympanic membrane (TM) thickness can provide crucial information for diagnosing several middle ear pathologies. An imaging system integrating low coherence interferometry (LCI) with the standard video otoscope has been shown as a promising tool for quantitative assessment of in-vivo TM thickness. The small field-of-view (FOV) of TM surface images acquired by the combined LCI-otoscope system, however, makes the spatial registration of the LCI imaging sites and their location on the TM difficult to achieve. It is therefore desirable to have a tool that can map the imaged points on to an anatomically accurate full-field surface image of the TM. To this end, we propose a novel automated mosaicking algorithm for generating a full-field surface image of the TM with co-registered LCI imaging sites from a sequence of multiple small FOV images and corresponding LCI data. Traditional image mosaicking techniques reported in the biomedical literature, mostly for retinal imaging, are not directly applicable to TM image mosaicking because unlike retinal images, which have several distinctive features, TM images contain large homogeneous areas lacking in sharp features. The proposed algorithm overcomes these challenges of TM image mosaicking by following a two-step approach. In the first step, a coarse registration based on the correlation of gross image features is performed. Subsequently, in the second step, the coarsely registered images are used to perform a finer intensity-based co-registration. The proposed algorithm is used to generate, for the first time, full-field thickness distribution maps of in-vivo human TMs.
|