Purpose: In the past years, great advances have been made in the way we diagnose various diseases. Most of the new approaches to medical diagnosis rely on imaging, and in large part imaging’s ability to compare different studies done at different times or from different modalities. however, finding anatomical or pathological changes over varying periods of time can still be difficult. Comparing studies can be quite challenging due to anatomical structure’s Complexity and changes in 3 dimensions that are hard to perceive. Using computational algorithms may lead to an improvement in the ability to compare between sets of images. In the current study an iterative study comparison algorithm was developed, The Algorithm includes a correlation in three planes and alignment between images shifted at different angles in the coronal and axial field of view.
Methods: An algorithm for matching non-contrast head CT studies was developed, the algorithm uses a multi-phase approach to matching studies. The first step of the algorithm organizes DICOM images into easily analyzed three-dimensional matrices. The algorithm creates a correlation on the x, y, and z-planes. The background of the images was removed using the Otsu thresholding method. The next step in analysis transforms axial images into synthetic coronal images to allow the algorithm to compute the error rate on that plane (Fig. 1). Angle corrections and comparisons of the studies is done iteratively until a minimal value of the mean square error between image series is reached. Using the algorithm analysis of 20 head CT scans of 10 patients (2 for each patient) were done and the mean square error was compared for each step of the algorithm.
Results: In the analysis of 10 series of two non contrast head CT studies showed a rate of mismatch between two different sets of images has decreased from an average of approximately 45% before analysis to 32% after the first step of iteration on the x, y and z planes. Adding the rotation correction algorithm with multiple iterations achieved a mismatch of less than 5% for all the compared scans.
Conclusions: Use of an iterative triple plane comparison algorithm is a feasible approach for comparison between two series of images of the same patient. Based on the results achieved thus far and with further improvements, this algorithm may be of future clinical use. Thus, Creating the basis for advanced study comparisons.
