Neuroimaging Software to Aid the Endovascular Surgeon: Automated Aneurysm Detection Using “Writhe” Number
Brain aneurysms, which are present in up to three percent of the general population, confront neurosurgeons with a difficult decision: preemptive treatment versus medical observation and serial follow-up. While elective aneurysm treatment continues to improve in both its endovascular and traditional microsurgical variants, it is still not free of risk and associated with direct and indirect costs to patients and the healthcare system.
A number of approaches have been proposed to determine the risk of any given aneurysm progressing toward rupture, a topic that continues to be an active area of research in medical imaging. As medical image analysis moves toward the mainstream, propelled in part by the exponential growth of digital imaging generated over the past decade, there exists an ever-increasing need to develop algorithms that are able to independently detect abnormal cerebral vessels and aneurysms.
Figure 1: Schematic of process flow used in writhe number- based aneurysm detection (from USPTO 8,781,194 B2 “Aneurysm Detection” by Malek, Miller and Lauric).
Figure 2: The writhe-based aneurysm detection algorithm can accept input from different imaging modalities providing three-dimensional volumetric datasets such as three-dimensional rotational angiography (3DRA), magnetic resonance angiography (MRA), cone-beam computed tomographic angiography (CBCTA), and conventional computed tomographic angiography (CTA).
The team of Adel Malek and Alex Lauric from the Cerebrovascular Hemodynamics Laboratory of Tufts Medical Center in Boston, Massachusetts, working with Eric Miller from the Department of Electrical Engineering at Tufts University, has been perfecting an algorithm based on a unique and novel approach.
Figure 3: Correct identification of a small basilar bifurcation aneurysm using the writhe number-based algorithm. Diagnostic cerebral angiogram (left panel); centerline in threedimensional space obtained from the three-dimensional rotational angiographic volume (center panel); red highlights the detected aneurysmal protrusion as detected by the algorithm (arrow, right panel).
This approach relies on an elegant mathematical descriptor called the writhe number. Initially introduced by F. Brock Fuller in 1971, the writhe number is used in curve theory to measure how a closed curve twists and coils around itself. It has been useful in characterizing the study and shape of DNA.
Our approach generalized writhe number from two-dimensional description to three-dimensional surface classification, and demonstrated that smooth tubular vessels that can be described by extrusion of a paraboloid should have a zero writhe value. The writhe-based aneurysm detection algorithm was recently awarded a U.S. patent, No. 8781194.
Our team then set out to use a non-zero writhe number to detect brain aneurysms on the surface of the three-dimensional network of brain vessels in patients with and without aneurysms. A number of advantages quickly became apparent when compared to existing aneurysm detection algorithms. Unlike other systems, the writhe-based approach does not require training to recognize aneurysms before it can begin detection. The writhe-based approach is also modality agnostic and has been successfully used on a range of 3D image datasets from different manufacturers. Aneurysms were successfully detected using the algorithm on 3D volume sets from computed tomographic angiography (CTA), magnetic resonance angiography (MRA), and three-dimensional rotational angiography (3DRA).
Coming advances in clinical non-invasive imaging in MRA include higher magnet strength and spatial resolution, which are expected to bring great benefit to the field and leverage writhe-based detection advantages over competing systems. The aneurysm detection algorithm can help clinicians when analyzing brain-imaging data, improve diagnostic accuracy at the time of interpretation, limit missed detection at the time of image acquisition, and work in the background to scour and scrub previously acquired studies to ensure no aneurysms were missed in the read.
Once an aneurysm is detected, additional shape characterization software can classify the lesion by location and determine its basic size in multiple dimensions such as height, width, and neck and parent vessel size to derive useful morphometric indices such as size, height-width, and aspect ratios. Aneurysm surface properties such as non-sphericity are then derived, paving the way to an estimation of propensity of progression towards rupture.