Abstract W MP48: Measurement of Intracranial Aneurysm Growth

Abstract

Purpose: The aim of this study is to develop a method to accurately measure 3-D growth of intracranial aneurysms using geodesic active contours segmentation and gradient descent registration algorithms. The method is expected to generate fast and reliable indication of aneurysm growth in clinical settings. Method: The geodesic active contours algorithm was implemented to perform 3-D aneurysm segmentation. 3-D CT Angiography images of 10 subjects with follow up aneurysm images were included in this study. Each subject had images acquired at two distinct time points (average time interval: 126.8 ± 80.1 days). 5 subjects were known to have aneurysm growth between the two time points (growth group). The other 5 subjects were known to have stable aneurysm sizes between the two time points (non-growth group). Segmentation and registration algorithms were used on images from all 10 subjects in an attempt to reproduce the diagnosis results. This study also included 7 sets of computer-built 3-D models and dimensions, for the purpose of validation. Result: Based on segmentation results, the growth group showed an average 21.7 ± 10.2% increase in aneurysm volume, and non-growth group showed an average 0.2 ± 1.3% increase in aneurysm volume. The difference in sample means between the growth group and non-growth group was statistically significant (p ≤ 0.05). For subjects in the growth group, the registration algorithm was able to identify and to label the growing region. For validation, segmentation and registration algorithms were also used on 7 sets of computer built models; growth was correctly labeled. The average runtime for processing segmentation under 4th Generation Intel Core i7-4500U Processor (3.0GHz) was 31.7 ± 38.4 seconds, and 45.2 ± 40.3 for registration. Input images had Signal-to-Noise Ratio (SNR) within a range of 1.8 to 5.9. Conclusion: Geodesic active contours segmentation and gradient descent registration offer an efficient way of measuring and monitoring intracranial aneurysm growth. Further improvements in algorithms are expected to reduce runtime and numerical errors.