SIIM: Researchers reveal intracranial aneurysm CAD
Researchers have developed a fully automatic computer-aided detection (CAD) scheme that can detect all shapes of intracranial aneurysms of different sizes in MR angiography (MRA) with high accuracy, according to research presented Thursday at the 2009 annual meeting of the Society for Imaging Informatics in Medicine (SIIM) held in Charlotte, N.C.
Xiaojiang Yang, PhD, of the Mayo Clinic in Rochester, Minn., and colleagues sought to develop an algorithm that could highlight features suggestive of aneurysms on 3D, intracranial time-of-flight (TOF) MRA exams, which are often used as an alternative to digital subtraction angiography (DSA).
Intracranial aneurysms can be categorized into three types based on morphology--saccular, bifurcation or fusiform--and size: big or small. Until now, none of the proposed methods could automatically detect all these types of aneurysms, according to the authors.
Yang and colleagues identified 58 3D TOF MRA studies in patients who had undergone intracranial DSA to confirm the presence of one or more aneurysms. The studies were then annotated by a trained radiologist to identify the aneurysm(s). The radiologist had access to reports and DSA images to increase confidence of findings.
The algorithm was executed on each of the 58 exams, and computed the number of true positives as those where points of interest were within 10 mm of the annotated aneurysm location. False positives were all the points of interest more than 10 mm from the annotation. No exact overlap was required.
Of the 58 randomly chosen data sets, 21 contained aneurysms, including all the three shapes, with a minimum size or 2 mm and maximum size 33 mm.
"Initially in our test, 18 aneurysms were successfully detected fully automatically, with an average false positives of 4.84 per exam. We found that the missing three aneurysms were due to the failure of segmentation that missed the aneurysms. If we manually adjusted the threshold used to segment the vessels in those three cases, all 21 aneurysms could be detected, with an average FPs 4.8," according to the researchers.
Yang and colleagues concluded that the addition of a CAD scheme that could highlight suspicious regions, while having a low FP rate, could improve the quality of healthcare provided at little or no increased radiologist effort.
"The result of the algorithm we developed reflects a high accuracy when vessels have been properly segmented. We believe not only that five FPs per case is in the acceptable range, but also that the number of FPs can be further reduced by combining some machine learning algorithms into the sieving process," the authors noted.
Xiaojiang Yang, PhD, of the Mayo Clinic in Rochester, Minn., and colleagues sought to develop an algorithm that could highlight features suggestive of aneurysms on 3D, intracranial time-of-flight (TOF) MRA exams, which are often used as an alternative to digital subtraction angiography (DSA).
Intracranial aneurysms can be categorized into three types based on morphology--saccular, bifurcation or fusiform--and size: big or small. Until now, none of the proposed methods could automatically detect all these types of aneurysms, according to the authors.
Yang and colleagues identified 58 3D TOF MRA studies in patients who had undergone intracranial DSA to confirm the presence of one or more aneurysms. The studies were then annotated by a trained radiologist to identify the aneurysm(s). The radiologist had access to reports and DSA images to increase confidence of findings.
The algorithm was executed on each of the 58 exams, and computed the number of true positives as those where points of interest were within 10 mm of the annotated aneurysm location. False positives were all the points of interest more than 10 mm from the annotation. No exact overlap was required.
Of the 58 randomly chosen data sets, 21 contained aneurysms, including all the three shapes, with a minimum size or 2 mm and maximum size 33 mm.
"Initially in our test, 18 aneurysms were successfully detected fully automatically, with an average false positives of 4.84 per exam. We found that the missing three aneurysms were due to the failure of segmentation that missed the aneurysms. If we manually adjusted the threshold used to segment the vessels in those three cases, all 21 aneurysms could be detected, with an average FPs 4.8," according to the researchers.
Yang and colleagues concluded that the addition of a CAD scheme that could highlight suspicious regions, while having a low FP rate, could improve the quality of healthcare provided at little or no increased radiologist effort.
"The result of the algorithm we developed reflects a high accuracy when vessels have been properly segmented. We believe not only that five FPs per case is in the acceptable range, but also that the number of FPs can be further reduced by combining some machine learning algorithms into the sieving process," the authors noted.