Fully automated detection and segmentation of intracranial aneurysms in subarachnoid hemorrhage on CTA using deep learning

Abstract

In aneurysmal subarachnoid hemorrhage (aSAH), accurate diagnosis of aneurysm is essential for subsequent treatment to prevent rebleeding. However, aneurysm detection proves to be challenging and time-consuming. The purpose of this study was to develop and evaluate a deep learning model (DLM) to automatically detect and segment aneurysms in patients with aSAH on computed tomography angiography. In this retrospective single-center study, three different DLMs were trained on 68 patients with 79 aneurysms treated for aSAH (2016-2017) using five-fold-cross-validation. Their outputs were combined to a single DLM via ensemble-learning. The DLM was evaluated on an independent test set consisting of 185 patients with 215 aneurysms (2010-2015). Independent manual segmentations of aneurysms in a 3D voxel-wise manner by two readers (neurosurgeon, radiologist) provided the reference standard. For aneurysms > 30 mm³ (mean diameter of ~ 4 mm) on the test set, the DLM provided a detection sensitivity of 87% with false positives (FPs)/scan of 0.42. Automatic segmentations achieved a median dice similarity coefficient (DSC) of 0.80 compared to the reference standard. Aneurysm location (anterior vs. posterior circulation; P = .07) and bleeding severity (Fisher grade ≤ 3 vs. 4; P = .33) did not impede detection sensitivity or segmentation performance. For aneurysms > 100 mm³ (mean diameter of ~ 6 mm), a sensitivity of 96% with DSC of 0.87 and FPs/scan of 0.14 were obtained. In the present study, we demonstrate that the proposed DLM detects and segments aneurysms > 30 mm³ in patients with aSAH with high sensitivity independent of cerebral circulation and bleeding severity while producing FP findings of less than one per scan. Hence, the DLM can potentially assist treating physicians in aSAH by providing automated detection and segmentations of aneurysms.

Figure 1

Flow chart for patient selection and inclusion scheme. aSAH = aneurysmal subarachnoid hemorrhage, CTA = CT-angiography.

Figure 2

Image preprocessing workflow in a patient with an aneurysm of the right middle cerebral artery (arrow) and Fisher 4 bleeding (intraventricular hemorrhage indicated by *) with mid-line shift. Fully automated brain mask computation, overlaid in green, from acquired head and neck CTA source images (a). Extracted brain image (b). 3D rendered computed vessel enhanced images with scale 0.5–5 voxels (c) and 5–15 voxels (d).

Figure 3

Performance of the different DLMs on the training set using five-fold-cross-validation (a). Segmentation performance of the DLM-Ens on the independent test set with respect to aneurysm volumes (b). Magenta circles represent the total number of false positives and green circles indicate the total number of false negatives.

Figure 4

Aneurysm volume correlation plots per patient between manually defined reference standard and the automatically obtained segmentations for validation (a) and test set (b) of DLM-Ens using Pearson correlation (r).

Figure 5

(a) Fifty-one-year-old male with Fisher 4 bleeding (* = intraventricular bleeding on axial unenhanced CT) and an aneurysm of the right middle cerebral artery (arrow) on axial CTA source images. The DLM-Ens (green) detects and segments the aneurysm (volume based on manual segmentation: 424.7 mm³) with high overlap (DSC of 0.94) compared to manual segmentations (red). (b) Sixty-three-year-old female with aSAH on axial unenhanced CT and an anterior communicating artery aneurysm (arrow) on axial CTA source images. Albeit being of small size (volume based on manual segmentation: 25.5 mm³), the DLM-Ens (green) detects and segments (DSC of 0.72) the aneurysm with high overlap compared to manual segmentations (red). (c) Fifty-one-year-old male with aSAH on axial unenhanced CT and a large basilar tip aneurysm (arrow; volume based on manual segmentation 419.6 mm³) on axial CTA source images. Compared to manual segmentations (red), the DLM-Ens (green) provides accurate detection and segmentation (DSC of 0.90).

Figure 6

(a) Axial CTA source images of a 64-year-old male with a small aneurysm (volume based on manual segmentation: 21.8 mm³) of the left internal carotid artery (arrow, red: manual segmentations) being missed by the DLM-Ens. (b) Axial CTA source images of a 65-year-old male with an anterior communicating artery aneurysm (arrow), which was accurately segmented and detected by the DLM. However, a false positive finding of the DLM-Ens (in green) at the superior sagittal sinus/transverse sinus was detected.