A hybrid technique using modified ICP algorithm for faster and automatic 2D & 3D microscopic image stitching in cytopathologic examination

Abstract

Due to the limitations of the light microscopic system such as limited depth of field and narrow field of view, entire sample areas are invisible and pathologists move the light microscope stage along the X -Y -Z axes with eye-hand coordination. In order to reduce the dependence on the pathologist and to allow whole sample areas to be examined in a short time without any control (without eye-hand coordination), this study creates 2D & 3D panoramic images with wide-view of sample in the light microscopic systems. According to our literature research, there is no study that creates 2D & 3D panoramic images in the microscopic system together. For this reason, it is thought that our study, which is the first one in which 2D & 3D panoramic images are created together in the light microscopic system, will be one of the pioneering studies in this field. Literature studies generally utilize specific microscope types such as laser, stereo and confocal, where 3D shape of the sample is extracted automatically and only 2D & 3D microscopic image stitching techniques are performed. Unlike these studies, due to the use of the light microscope which is requires a procedure to extract the 3D structure of the sample, this study contains an extra phase which creates 2D & 3D adjacent images using series of multi-focus 2D adjacent images before 2D & 3D microscopic image stitching technique. Moreover, a hybrid 2D & 3D microscopic image stitching technique which uses modified ıterative closest point algorithm and combines the stages of 2D and 3D image-based 2D & 3D image stitching techniques is developed in this study. Both qualitative and quantitative evaluations of the proposed study are performed on real microscope image data sets from samples prepared for cytopathologic examination in the light microscopic system. Our hybrid technique creates 2D panoramic images whose values of peak signal to noise ratio (PSNR), correlation coefficient (CC), entropy and average gradient are computed respectively as 30.4267, 0.9997, 7.4670 and 122.2510 and 3D panoramic images whose values of PSNR, universal quality index, root mean square error, CC, kurtosis metric and standard deviation are computed respectively as 17.7458, 0.9702, 8.9230, 0.8702, 5.4981 and 31.006, which are the best values when compared with well-established studies. Moreover, the proposed hybrid technique with 9.8403 and 23.5902 (s) execution times accelerates the process of automatic 2D & 3D panoramic imaging.