||6th Annual ESHE Meeting
||Diez-Pastor, José-Francisco; García González, Rebeca; Chico Carrancio, Sergio; Modesto-Mata, Mario; Marticorena Sánchez, Raúl
Perikymata are the surface manifestations of Retzius line, which show biological rhythms range from 6 to 12 days in hominoids. Number and distribution of these long-period lines provide insights in taxonomical diagnosis in hominins, as well as allow us to obtain an accurate enough estimation of crown formation time. Perikymata are generally visualized on high-resolution images obtained in different microscopes using a magnification of 50X and counts are typically made manually. This procedure is high time consuming and could be a source of inter-observer error. For this reason, here we present a novel semi-automatic method to obtain, both, the number and the distribution of perikymata. In general the proposed method can be split in two main stages: the first one consists of reconstructing the image of the tooth from multiple fragments or subimages that are obtained by the microscope and, in the second stage, the system detects and analyzes the distribution of perikymata. The designed method takes the set of images obtained by the microscope, given the resolution of these devices, a single tooth can be represented by more than a dozen of images, each one showing a small part of it. Panoramic image reconstruction is a problem that is similar to solving puzzles; this problem has been widely addressed in digital photography research. In the first place, the image reconstruction algorithm detects invariant features within image fragments, secondly the spatial relations between imagefragments are found through the matching of their invariant features, and finally transformations like scaling, rotations or color compensation are performed to produce the final image. This stage of the method eliminates the need for manual processing using image editing tools, what it is very time consuming. In the next stage of the method, starting from a full reconstructed image, the researcher should delimit the work area of the image to be processed. This area marks the upper and lower bound of the crown. This work area is automatically divided into deciles (ten equal parts). Often, the perikymata look blurred, fuzzy or eroded in the image and they are hard to recognize, the researcher must draw lines that cross the perikymata by regions where they can be easily distinguished. Using these lines as guides, it is performed automatically the counting and analysis of perikymata. Intuitively the perikymata can be defined on an image as lines of lower intensity flanked by lines of greater intensity. Using the intensities of the lines drawn by the researcher an intensity profile is built. The problem boils down to deciding which of the valleys in this intensity profile correspond to perikymata. To analyze this profile intensities we have tested edge detection algorithms, function approximation using neural networks or combinatorial optimization, to determine which are the most likely points defining perikymata. After that, the researcher can validate these perikymata detected by the tool or can correct some of these detections. After validation by the researcher, calculating the distances between perikymata and distribution in deciles it is done automatically. In our opinion this semi-automatic method involves a great improvement over the manual procedure performed nowadays, it eliminates errors caused by fatigue or lack of concentration, and therefore, obtaining better results and streamlining the tasks of the researcher.