Use of a convolutional neural network to identify infarct core using computed tomography perfusion parameters

Abstract

Purpose: Computed tomography perfusion (CTP) is used to diagnose ischemic strokes through contralateral hemisphere comparisons of various perfusion parameters. Various perfusion parameter thresholds have been utilized to segment infarct tissue due to differences in CTP software and patient baseline hemodynamics. This study utilized a convolutional neural network (CNN) to eliminate the need for non-universal parameter thresholds to segment infarct tissue.

Methods: CTP data from 63 ischemic stroke patients was retrospectively collected and perfusion parameter maps were generated using Vitrea CTP software. Infarct ground truth labels were segmented from diffusion-weighted imaging (DWI) and CTP and DWI volumes were registered. A U-net based CNN was trained and tested five separate times using each CTP parameter (cerebral blood flow (CBF), cerebral blood volume (CBV), time-to-peak (TTP), mean-transit-time (MTT), delay time). 8,352 infarct slices were utilized with a 60:30:10 training:testing:validation split and Monte Carlo cross-validation was conducted using 20 iterations. Infarct volumes were reconstructed following segmentation from each CTP slice. Infarct spatial and volumetric agreement was compared between each CTP parameter and DWI.

Results: Spatial agreement metrics (Dice coefficient, positive predictive value) for each CTP parameter in predicting infarct volumes are: CBF=(0.67, 0.76), CBV=(0.44, 0.62), TTP=(0.60, 0.67), MTT=(0.58, 0.62), delay time=(0.57, 0.60). 95% confidence intervals for volume differences with DWI infarct are: CBF=14.3±11.5 mL, CBV=29.6±21.2 mL, TTP=7.7±15.2 mL, MTT=-10.7±18.6 mL, delay time=-5.7±23.6 mL.

Conclusions: CBF is the most accurate CTP parameter in segmenting infarct tissue. Segmentation of infarct using a CNN has the potential to eliminate non-universal CTP contralateral hemisphere comparison thresholds.

Keywords: Computed tomography perfusion; cerebral infarct tissue; convolutional neural network; semantic segmentation.

Figure 1:

Represents a CBF CTP map from a patient with a middle cerebral artery occlusion. The color bar on the left indicates a maximum CBF of 66 mL/100g/min and a minimum of 0 mL/100g/min.

Figure 2:

Demonstrates a TDC and how the CTP parameters are extracted from the curve. Delay time is excluded from the curve but represents the time at which the curve intensity goes above 0.¹⁰

Figure 3:

Image (a) showing an example of a CTP CBF slice that is fed into the network, (b) the DWI slice corresponding the CTP slice, and (c) indicates the ground truth infarct label extracted through automated segmentation from the DWI. Infarct positive and negative regions in (c) are indicated by pixel values of 1 and 0, respectively.

Figure 4:

Represents a slice from a CTP scan (a) following registration with a DWI volume whose corresponding slice is indicated in (b). Image (c) shows the spatial overlap of the ventricles from the two images with the CTP ventricles indicates in green, the DWI ventricle in pink, and the overlap of the two in white.

Figure 5:

Modified U-net architecture utilized for segmentation of infarct from CTP parameter maps.

Figure 6:

Indicates a CBF perfusion slice (a), the ground truth infarct label from DWI (b), and the predicted infarct region from the CBF perfusion map (c). Ground truth infarct from DWI is indicated by the red outlines in (a) and (c).

Figure 7:

Indicates a perfusion slice from the CBV map (a), the corresponding ground truth infarct label from DWI (b), and the CNN predicted infarct region based on the perfusion map (c). Ground truth infarct from DWI is indicated by the red outlines in (a) and (c). Note the erroneous infarct prediction in the contralateral hemisphere in (c).

Figure 8:

Shows a failed segmentation of infarct within the posterior cerebral artery territory, (a) indicates the CBF perfusion map fed into the CNN for infarct prediction, (b) indicates the infarct ground truth location from DWI, and (c) indicates the prediction of infarct which is blank. Ground truth infarct from DWI is indicated by the red outlines in (a) and (c).

Figure 9:

Indicates the reconstructed CNN predicted CTP infarct volumes overlapping with the final infarct volumes from DWI. Maroon regions indicate infarct overlap between the two modalities, blue regions are infarct from just DWI, and green regions are CNN CTP predicted infarct.

Figure 10:

Shows the reconstructed watershed corrected CNN predicted CTP infarct volumes overlapping with the final infarct volumes from DWI. Maroon regions indicate infarct overlap between the two modalities, blue regions are infarct from just DWI, and green regions are CNN CTP predicted infarct. Note the elimination of erroneous infarct in the contralateral hemispheres.