Publication

Abstract : This study aims to assess improved methods of image registration to obtain accurate alignment of CT scans of the liver and brain. A collection of more than 1600 CT scans of the brain and liver is used in our research. We employ a variety of techniques, such as FFT-based registration with frequency domain analysis, Cv2 (OpenCV) for complete computer vision capabilities including picture alignment, and PystackReg, a Python package with a variety of registration algorithms. While Cv2 enables feature detection and geometric transformations, PystackReg offers a variety of transformation models. FFT uses spectral shifts to align images by taking advantage of frequency domains. By addressing particular aspects of the image, these techniques allow for accurate alignment of CT images throughout the registration process. As our CT dataset consists of two different anatomic data (Brain and Liver), we will divide the outcomes of each approach into two sections. To conclude our findings, we employ a variety of evaluation measures, including the Euclidean distance, Mean Square Error, Sum of Absolute Differences, and Jaccard Index. The FFT model is found to be more accurate in the dataset including brain CT scans, as evidenced by its higher Jaccard Index value (0.909) in comparison to the other models (0.439-PystackReg, 0.866-Cv2) and its lower Mean Square Error (0.090) in comparison to the other models (37.708-PystackReg, 0.130-Cv2). However, with the lowest Euclidean Measure (0.5974) and mean square error (0.0050) as compared to PystackReg (14.9810, 0.0386) and FFT (5.3996, 0.5727) models, respectively, the Cv2 Model performs better on liver CT images. These advancements could lead to better therapeutic evaluation, diagnostic capabilities, and personalized healthcare. They might also bring about a paradigm shift towards the analysis and interpretation of medical pictures that is more successful and efficient.