Our Work
Neuroimaging of the developing (perinatal) brain challenges radiological interpretation due to the rapid changes seen over relatively short timeframes (weeks). Both expected developmental changes in intensity, shape and size and variety occurring during this period. This PIPPI challenge offer the community a unique opportunity to explore three key tasks that address heterogeneity of age and injury seen in neonatal imaging. Using the large and openly shared developing Human Connectome (dHCP) dateset, teams will develop a series of tools that will be tested on unseen hold out data by the organisers. This hold-out data will include images acquired in a similar fashion to the dHCP but also images acquired as part of other studies, which may have different MRI acquisition parameters.
Challenge 01
Gestational Age Prediction
The neonatal brain develops rapidly in terms of size, shape and contrast over periods of days and weeks. Though this development is highly stereotyped across infants, being born prematurely can interrupt normal processes (such as myelination) in addition to confer risks to subtle brain injuries. Using brain imaging data collected over the perinatal period, as part of the developing Human Connectome Project (dHCP), this challenge will ask entrants to guess the age at which a neonate was born (gestational age at birth, this will include infants born prematurely) irrespective of the age at scan (which may be any time from 28 to 44 gestational weeks). The resulting model will be applied to a held out test dataset from two sites (one included in the training data, one unseen).
Challenge 02
Abnormality Detection
Brain abnormalities and injuries seen in neonates (such punctate white matter lesions or small haemorrhages) are heterogeneous in location and cause. This, in addition to the rapid expected and normal changes seen in the brain, makes radiological detection and quantification of injury challenging. Methods to detect tissue abnormalities automatically are highly biased towards training on adult-age imaging data, which is relatively stable in terms of contrast for a given acquisition. This challenge will aim to detect manually delineated tissue abnormalities in a cohort of infants collected over a wide range of ages (28 to 44 weeks post-menstrual age) as part of the developing Human Connectome Project (dHCP). The resulting model or classification tool will then be used to detect injury in entirely held out data (a) collected in the same way as the dHCP and (b) collected with a legacy, lower resolution dataset.
Images
Figures display T2w images of preterm neonates with injuries and their corresponding injury masks.
Fig 1: 32 week old female neonate with germinal matrix hemorrhage, inter-ventricular hemorrhagic and hemorrhagic parenchymal infarct.
Fig 2: 35 week old female neonate with periventricular leukomalacia.
Fig 3: 33 week old female neonate with multiple florid punctate white matter lesions.
Figure 1
Figure 2
Figure 3
