What to Look for on Post-stroke Neuroimaging

Unsupervised Deep Learning for Stroke Lesion Segmentation on Follow-up CT Based on Generative Adversarial Networks

BACKGROUND AND PURPOSE

Supervised deep learning is the state-of-the-art method for stroke lesion segmentation on NCCT. Supervised methods require manual lesion annotations for model development, while unsupervised deep learning methods such as generative adversarial networks do not. The aim of this study was to develop and evaluate a generative adversarial network to segment infarct and hemorrhagic stroke lesions on follow-up NCCT scans.

MATERIALS AND METHODS

Training data consisted of 820 patients with baseline and follow-up NCCT from 3 Dutch acute ischemic stroke trials. A generative adversarial network was optimized to transform a follow-up scan with a lesion to a generated baseline scan without a lesion by generating a difference map that was subtracted from the follow-up scan. The generated difference map was used to automatically extract lesion segmentations. Segmentation of primary hemorrhagic lesions, hemorrhagic transformation of ischemic stroke, and 24-hour and 1-week follow-up infarct lesions were evaluated relative to expert annotations with the Dice similarity coefficient, Bland-Altman analysis, and intraclass correlation coefficient.

RESULTS

The median Dice similarity coefficient was 0.31 (interquartile range, 0.08-0.59) and 0.59 (interquartile range, 0.29-0.74) for the 24-hour and 1-week infarct lesions, respectively. A much lower Dice similarity coefficient was measured for hemorrhagic transformation (median, 0.02; interquartile range, 0-0.14) and primary hemorrhage lesions (median, 0.08; interquartile range, 0.01-0.35). Predicted lesion volume and the intraclass correlation coefficient were good for the 24-hour (bias, 3 mL; limits of agreement, -64-59 mL; intraclass correlation coefficient, 0.83; 95% CI, 0.78-0.88) and excellent for the 1-week (bias, -4 m; limits of agreement,-66-58 mL; intraclass correlation coefficient, 0.90; 95% CI, 0.83-0.93) follow-up infarct lesions.

CONCLUSIONS

An unsupervised generative adversarial network can be used to obtain automated infarct lesion segmentations with a moderate Dice similarity coefficient and good volumetric correspondence.

Brain Magnetic Resonance Imaging Volumetric Measures of Functional Outcome after Severe Traumatic Brain Injury in Adolescents

Adolescent traumatic brain injury (TBI) is a major public health concern, resulting in >35,000 hospitalizations in the United States each year. Although neuroimaging is a primary diagnostic tool in the clinical assessment of TBI, our understanding of how specific neuroimaging findings relate to outcome remains limited. Our study aims to identify imaging biomarkers of long-term neurocognitive outcome after severe adolescent TBI. Twenty-four adolescents with severe TBI (Glasgow Coma Scale ≤8) enrolled in the ADAPT (Approaches and Decisions after Pediatric TBI) study were recruited for magnetic resonance imaging (MRI) scanning 1-2 years post-injury at 13 participating sites. Subjects underwent outcome assessments ∼1-year post-injury, including the Wechsler Abbreviated Scale of Intelligence (IQ) and the Pediatric Glasgow Outcome Scale-Extended (GOSE-Peds). A typically developing control cohort of 38 age-matched adolescents also underwent scanning and neurocognitive assessment. Brain-image segmentation was performed on T1-weighted images using Freesurfer. Brain and ventricular cerebrospinal fluid volumes were used to compute a ventricle-to-brain ratio (VBR) for each subject, and the corpus callosum cross-sectional area was determined in the midline for each subject. The TBI group demonstrated higher VBR and lower corpus callosum area compared to the control cohort. After adjusting for age and sex, VBR was significantly related with GOSE-Peds score in the TBI group (n = 24, p = 0.01, cumulative odds ratio = 2.18). After adjusting for age, sex, intracranial volume, and brain volume, corpus callosum cross-sectional area correlated significantly with IQ score in the TBI group (partial cor = 0.68, n = 18, p = 0.007) and with PSI (partial cor = 0.33, p = 0.02). No association was found between VBR and IQ or between corpus callosum and GOSE-Peds. After severe adolescent TBI, quantitative MRI measures of VBR and corpus callosum cross-sectional area are associated with global functional outcome and neurocognitive outcomes, respectively.

Conflicting clinical and radiological management decisions

Gliosis with hemorrhagic transformation is a late reported complication of stroke. Sometimes there is a big discrepancy between clinical and radiological diagnosis, and clinical decisions must be multi-aspect decisions and not dependent on a single discrepant investigation result.