Neonatal Pain and Infection Relate to Smaller Cerebellum in Very Preterm Children at School Age.

OBJECTIVE: To examine whether specific neonatal factors differentially influence cerebellar subregional volumes and to investigate relationships between subregional volumes and outcomes in very preterm children at 7 years of age. STUDY DESIGN: Fifty-six children born very preterm (24-32 weeks gestational age) followed longitudinally from birth underwent 3-dimensional T(1)-weighted neuroimaging at median age 7.6 years. Children with severe brain injury were excluded. Cerebellar subregions were automatically segmented using the multiple automatically generated templates algorithm. The relation between cerebellum subregional volumes (adjusted for total brain volume and sex) and neonatal clinical factors were examined using constrained principal component analysis. Cognitive and visual-motor integration functions in relation to cerebellar volumes were also investigated. RESULTS: Higher neonatal procedural pain and infection, as well as other clinical factors, were differentially associated with reduced cerebellar volumes in specific subregions. After adjusting for clinical risk factors, neonatal procedural pain was distinctively associated with smaller volumes bilaterally in the posterior VIIIA and VIIIB lobules. Specific smaller cerebellar subregional volumes were related to poorer cognition and motor/visual integration. CONCLUSIONS: In very preterm children, exposure to painful procedures, as well as additional neonatal risk factors such as infection, were associated with reduced cerebellar volumes in specific subregions and poorer outcomes at school age.

Propagating Uncertainty Across Cascaded Medical Imaging Tasks for Improved Deep Learning Inference.

Although deep networks have been shown to perform very well on a variety of medical imaging tasks, inference in the presence of pathology presents several challenges to common models. These challenges impede the integration of deep learning models into real clinical workflows, where the customary process of cascading deterministic outputs from a sequence of image-based inference steps (e.g. registration, segmentation) generally leads to an accumulation of errors that impacts the accuracy of downstream inference tasks. In this paper, we propose that by embedding uncertainty estimates across cascaded inference tasks, performance on the downstream inference tasks should be improved. We demonstrate the effectiveness of the proposed approach in three different clinical contexts: (i) We demonstrate that by propagating T2 weighted lesion segmentation results and their associated uncertainties, subsequent T2 lesion detection performance is improved when evaluated on a proprietary large-scale, multi-site, clinical trial dataset acquired from patients with Multiple Sclerosis. (ii) We show an improvement in brain tumour segmentation performance when the uncertainty map associated with a synthesised missing MR volume is provided as an additional input to a follow-up brain tumour segmentation network, when evaluated on the publicly available BraTS-2018 dataset. (iii) We show that by propagating uncertainties from a voxel-level hippocampus segmentation task, the subsequent regression of the Alzheimer's disease clinical score is improved.