Analysis and evaluation of the calibration of 2D-3D medical image registration clinical datasets
Objective Constructing 2D-3D medical image registration datasets using clinical materials is important for various learning algorithms to be applied in medical treatment.The dataset's calibration needs to be analyzed and evaluated because of various uncertainties in the acquisition process of clinical data.In the paper,several sets of calibration values between a group of X-ray images and CT image of a patient in thoracic endovascular aortic repair were analyzed and evaluated,and the calibration was determined finally.Methods The calibration values were analyzed and evaluated by similarity metric and projection distance error,respectively.The similarity criterion was used to calculate the similarity between the 2D Digitally reconstructed radiograph(DDR)generated from the CT image and the X-ray image.The higher the similarity,the closer the corresponding calibration value was to the true value.After reading the markers'position in the X-ray image as reference position,and obtaining projection position by projecting the markers'positions calculated in the CT image,the distance between the reference position and the projected position,projection distance error of the markers,was calculated.The smaller the distance error,the closer the corresponding calibration value was to the true value.Results The calculations for several sets of calibration values had no significant difference in comparing the similarity between DRRs and X-ray images,whereas the calculations of the projected distance errors had obvious difference and were able to quantitatively describe the advantages and disadvantages of the calibrations.The main reason was that the calibration values of each group were close,and mode difference between the DRR images and X-ray images was significant.Conclusions It can be concluded that the projected distance error evaluation is an effective means for evaluating the calibrations of 2D-3D medical image registration datasets;in addition,the similarity metric evaluation is also an effective way of evaluating the calibrations if the difference of the calibrations provided is obvious,or high-quality DRRs can be provided.
万方数据
维普
