Sickle Cell Anemia Impacts Brain Age in Children

Children with sickle cell anemia (SCA) are at risk for cognitive dysfunction independent of infarction, suggesting SCA impacts brain function and brain development prior to stroke occurrence. Research-only imaging sequences demonstrate microstructural differences in children with SCA. However, these sequences are not available clinically and require processing expertise. DeepBrainNet is a convolutional neural network that accurately predicts brain age across the lifespan using a single, clinically available T1 sequence (PMID 32591831). Both predicted age and brain age gap (BAG; difference between predicted brain age and chronological age) are reliable indicators of brain maturation (PMID 34892076). We utilized DeepBrainNet to calculate brain age in children with SCA and healthy controls (HC) to test the hypothesis that children with SCA would have delayed brain maturation compared to HC.

Brain MRIs were collected longitudinally with laboratory evaluation from HC and children with SCA between 5-21 years of age. We excluded those with cerebral vasculopathy on MRA, history of stem cell transplant or gene therapy, disorders other than SCA associated with neurocognitive complications, and contraindications to MRI. MRIs were skull stripped, affinely registered to atlas space, and split into 80 axial slices. Each slice ran through DeepBrainNet independently, resulting in 80 brain age predictions per T1 image. Whole brain (WB) predicted age is the median of these 80 values (PMID 32591831).

DeepBrainNet predicted brain age from 126 brain MRIs (34 HC, 92 SCA) acquired from 102 participants (27 HC, 75 SCA). HC and SCA cohorts were matched for age (HC 13.5 (±3.3) years, SCA 12.9 (±4.2) years, p = 0.498) and sex (p = 0.455) upon enrollment. Twenty-nine (38.7%) SCA participants had a history of silent infarcts (SCI); 5 (6.7%) had a history of overt stroke; 49 (65.3%) received hydroxyurea and 20 (26.7%) received chronic transfusion therapy. The average BAG was 3.2 (±1.9) years for HC and 2.6 (±2.6) years for SCA, indicating that DeepBrainNet overestimated brain age for all participants, including HC. We accounted for the previously reported bias of overestimating brain age of young participants (PMID 32120292) by including chronological age in all statistical models. With linear mixed effects modeling to control for repeated measures and chronological age, there was a significant difference in WB BAG between HC (3.5 (95% CI 2.7, 4.3) years) and SCA without SCI or overt stroke (2.4 (95% CI 1.8, 3.0) years, p = 0.038). Considering all SCA participants, there was not a difference in WB BAG between HC (3.4 (95% CI 2.5, 4.4) years) and SCA (2.6 (95% CI 2.0, 3.1) years, p=0.106). Consistent with our finding of higher BAG in HC, BAG increased by 0.19 (95% CI -0.02, 0.40) years for each 1 g/dL increase in hemoglobin, but this did not reach significance (p = 0.071). The slices within DeepBrainNet corresponding to the region at greatest risk for SCI (PMID 30061156) were identified. The predicted brain age in these high-risk slices had a larger group difference in BAG (SCA-HC, -1.3 (95% CI -2.5, -0.1)) versus the remainder of the brain (-0.7 (95% CI -1.9, 0.5)), with significance achieved for an interaction term between region at greatest risk of SCI and cohort (p = 0.003).

WB BAG in SCA without stroke is significantly lower than HC, suggesting that there is delayed brain maturation in children with SCA compared to HC. Furthermore, a previously defined high-risk region that is vulnerable to SCI due to hemodynamic compromise shows a larger group difference in BAG compared to the remainder of the brain, suggesting that the differences in WB BAG may be driven by a regionally specific effect of SCA. While this specific high-risk region is at risk for infarction, the WB findings suggest differences are not entirely driven by stroke. The lack of difference in BAG between HC and SCA when participants with stroke were included suggests that stroke increases brain age prediction with DeepBrainNet, potentially obscuring other structural changes present prior to stroke. Requiring only clinically available MRI sequences, these data suggest that DeepBrainNet has potential to become a biomarker for overall brain health in children with SCA with continued refinement. Further investigation is required to understand the trajectory of BAG with aging in SCA, impact of therapeutics on BAG and relationship between BAG and neurocognitive outcomes.