Optimizing image registration and infarct definition in stroke research

Abstract

ObjectiveAccurate representation of final infarct volume is essential for assessing the efficacy of stroke interventions in imaging‐based studies. This study defines the impact of image registration methods used at different timepoints following stroke, and the implications for infarct definition in stroke research.MethodsPatients presenting with acute ischemic stroke were imaged serially using magnetic resonance imaging. Infarct volume was defined manually using four metrics: 24‐h b1000 imaging; 1‐week and 1‐month T2‐weighted FLAIR; and automatically using predefined thresholds of ADC at 24 h. Infarct overlap statistics and volumes were compared across timepoints following both rigid body and nonlinear image registration to the presenting MRI. The effect of nonlinear registration on a hypothetical trial sample size was calculated.ResultsThirty‐seven patients were included. Nonlinear registration improved infarct overlap statistics and consistency of total infarct volumes across timepoints, and reduced infarct volumes by 4.0 mL (13.1%) and 7.1 mL (18.2%) at 24 h and 1 week, respectively, compared to rigid body registration. Infarct volume at 24 h, defined using a predetermined ADC threshold, was less sensitive to infarction than b1000 imaging. 1‐week T2‐weighted FLAIR imaging was the most accurate representation of final infarct volume. Nonlinear registration reduced hypothetical trial sample size, independent of infarct volume, by an average of 13%.InterpretationNonlinear image registration may offer the opportunity of improving the accuracy of infarct definition in serial imaging studies compared to rigid body registration, helping to overcome the challenges of anatomical distortions at subacute timepoints, and reducing sample size for imaging‐based clinical trials.

Quantifying Infarct Growth and Secondary Injury Volumes

“…The same registration-based approach could also be implemented in preclinical stroke models, facilitating translation of novel approaches into early phase trials 34. More consistent measurements improve statistical power in such studies, offering the potential to reduce sample size 20…”

Quantifying Infarct Growth and Secondary Injury Volumes

“…The same registration-based approach could also be implemented in preclinical stroke models, facilitating translation of novel approaches into early phase trials 34. More consistent measurements improve statistical power in such studies, offering the potential to reduce sample size 20…”

Cerebral blood flow quantification with multi-delay arterial spin labeling in ischemic stroke and the association with early neurological outcome

“…GM masks from the MNI standard atlas registered to ASL image space by inverse normalization. In order to do so, we first performed linear registration of M0 images to FLAIR images followed by non-linear registration (FNIRT) of FLAIR images to a FLAIR-based MNI standard atlas in order to match relative signal intensities between the input and reference images ( Winkler et al, 2009 , Harston et al, 2017 ). Tissue masks with a Tmax delay >6 s (smoothed 5 mm FWHM) were registered to ASL space by performing rigid registration of CTP images to M0 images using a mutual cost function and without prior brain extraction (BET) as this has been shown to result in more accurate intermodal registration between CT and MRI ( Hinds et al, 2018 ).…”

Defining the optimal target for anterior thalamic deep brain stimulation in patients with drug-refractory epilepsy

“…13 Previous studies have shown that nonlinear models provide greater accuracy over automated linear methods as well as manual registration methods by accommodating differences in brain morphology. 29,30 Thus, we could obtain an accurate lead localization that used nonlinear registrations to better account for the intersubject morphological variations and allow for convenient intersubject analysis.…”

Quantifying Infarct Growth and Secondary Injury Volumes

“…The same registration-based approach could also be implemented in preclinical stroke models, facilitating translation of novel approaches into early phase trials 34. More consistent measurements improve statistical power in such studies, offering the potential to reduce sample size 20…”

Cerebral blood flow quantification with multi-delay arterial spin labeling in ischemic stroke and the association with early neurological outcome

“…GM masks from the MNI standard atlas registered to ASL image space by inverse normalization. In order to do so, we first performed linear registration of M0 images to FLAIR images followed by non-linear registration (FNIRT) of FLAIR images to a FLAIR-based MNI standard atlas in order to match relative signal intensities between the input and reference images ( Winkler et al, 2009 , Harston et al, 2017 ). Tissue masks with a Tmax delay >6 s (smoothed 5 mm FWHM) were registered to ASL space by performing rigid registration of CTP images to M0 images using a mutual cost function and without prior brain extraction (BET) as this has been shown to result in more accurate intermodal registration between CT and MRI ( Hinds et al, 2018 ).…”

Defining the optimal target for anterior thalamic deep brain stimulation in patients with drug-refractory epilepsy

“…13 Previous studies have shown that nonlinear models provide greater accuracy over automated linear methods as well as manual registration methods by accommodating differences in brain morphology. 29,30 Thus, we could obtain an accurate lead localization that used nonlinear registrations to better account for the intersubject morphological variations and allow for convenient intersubject analysis.…”

Quantifying Infarct Growth and Secondary Injury Volumes

“…The same registration-based approach could also be implemented in preclinical stroke models, facilitating translation of novel approaches into early phase trials 34. More consistent measurements improve statistical power in such studies, offering the potential to reduce sample size 20…”

Cerebral blood flow quantification with multi-delay arterial spin labeling in ischemic stroke and the association with early neurological outcome

“…GM masks from the MNI standard atlas registered to ASL image space by inverse normalization. In order to do so, we first performed linear registration of M0 images to FLAIR images followed by non-linear registration (FNIRT) of FLAIR images to a FLAIR-based MNI standard atlas in order to match relative signal intensities between the input and reference images ( Winkler et al, 2009 , Harston et al, 2017 ). Tissue masks with a Tmax delay >6 s (smoothed 5 mm FWHM) were registered to ASL space by performing rigid registration of CTP images to M0 images using a mutual cost function and without prior brain extraction (BET) as this has been shown to result in more accurate intermodal registration between CT and MRI ( Hinds et al, 2018 ).…”

Defining the optimal target for anterior thalamic deep brain stimulation in patients with drug-refractory epilepsy

“…13 Previous studies have shown that nonlinear models provide greater accuracy over automated linear methods as well as manual registration methods by accommodating differences in brain morphology. 29,30 Thus, we could obtain an accurate lead localization that used nonlinear registrations to better account for the intersubject morphological variations and allow for convenient intersubject analysis.…”