Streamlining Acute Abdominal Aortic Dissection Management-An AI-based CT Imaging Workflow

Abstract

Life-threatening acute aortic dissection (AD) demands timely diagnosis for effective intervention. To streamline intrahospital workflows, automated detection of AD in abdominal computed tomography (CT) scans seems useful to assist humans. We aimed at creating a robust convolutional neural network (CNN)-based pipeline capable of real-time screening for signs of abdominal AD in CT. In this retrospective study, abdominal CT data from AD patients presenting with AD and from non-AD patients were collected (n 195, AD cases 94, mean age 65.9 years, female ratio 35.8%). A CNN-based algorithm was developed with the goal of enabling a robust, automated, and highly sensitive detection of abdominal AD. Two sets from internal (n = 32, AD cases 16) and external sources (n = 1189, AD cases 100) were procured for validation. The abdominal region was extracted, followed by the automatic isolation of the aorta region of interest (ROI) and highlighting of the membrane via edge extraction, followed by classification of the aortic ROI as dissected/healthy. A fivefold cross-validation was employed on the internal set, and an ensemble of the 5 trained models was used to predict the internal and external validation set. Evaluation metrics included receiver operating characteristic curve (AUC) and balanced accuracy. The AUC, balanced accuracy, and sensitivity scores of the internal dataset were 0.932 (CI 0.891-0.963), 0.860, and 0.885, respectively. For the internal validation dataset, the AUC, balanced accuracy, and sensitivity scores were 0.887 (CI 0.732-0.988), 0.781, and 0.875, respectively. Furthermore, for the external validation dataset, AUC, balanced accuracy, and sensitivity scores were 0.993 (CI 0.918-0.994), 0.933, and 1.000, respectively. The proposed automated pipeline could assist humans in expediting acute aortic dissection management when integrated into clinical workflows.

Keywords: Abdomen; Aortic dissection; Computed tomography; Convolutional neural network; Deep learning.

Fig. 1

Example images from internal and external datasets. The left column is healthy cases and the right column is aortic dissection (AD) cases

Fig. 2

CT volume preprocessing pipeline. The abdominal region is automatically extracted and resized. An aorta mask is generated from this region and then dilated to mask out the aorta region of interest (ROI) from the abdomen region. For highlighting aorta bifurcation in AD cases, edges are extracted using the canny detector. The aorta ROI is weighted down by a factor of 0.6 where there is no edge voxel present. Finally, the edge-weighted aorta ROI is standardized by cropping/padding with the bounding box of the dilated aortic mask

Fig. 3

Network architecture for aortic dissection classification. The input volume (edge-weighted aorta ROI) is processed with a CNN to produce a single output in the range [0, 1]. The network consists of convolutional blocks with convolutions of size three, followed by instance norm, dropout, ReLU activation, and max-pool of size two (except in the last convolution block (orange block)). The final feature vector is produced by an average-pooling. It is then processed by a dense layer and a sigmoid activation to produce the output

Fig. 4

Internal dataset patient selection flowchart. Data acquisition process. The healthy controls (n = 101) were collected similarly to the included AD cases (n = 94)

Fig. 5

AUC curves for internal cross validation (CV), internal validation (valid), and external sets. The internal set AUC is 0.93, internal validation set AUC is 0.89, while the external set AUC is 0.99