Data Description¶

Features¶

The challenge data set will include the same data sets and splits as used by [Fawaz et al., 2021; Dahan et al., 2022]: these include cortical surface meshes and 4 cortical metric files (sulcal depth, cortical thickness, curvature and T1/T2 myelin maps; shown below) acquired from 514 neonates scanned as part of the dHCP project, of which 95 are born preterm (\<37 weeks’ gestational age). We have also included white and pial surfaces, which participants might find useful for the derivation of other morphological features such as surface area.

An example script for resampling the cortical surfaces and metrics to a regularly-tessellated sixth-order icosphere is included alongside the data.

Metadata/Covariates¶

Age at scan
The prediction of birth age is confounded by the age at scan (postmenstrual age), as not all babies are scanned at exactly the same time.  Therefore, we provide age at scan as additional metadata for participants, and exploration of different methods of deconfounding methods on model performance are encouraged.

Biological sex
Neither [Fawaz et al., 2021; Dahan et al., 2022] included biological sex as an additional confound in predicting birth age. However, it is well documented that differences in brain development do exist between biological females and males during childhood [Kaczkurkin et al., 2018], and  participants may want to also include biological sex as an additional confound in predicting age at birth or neurodevelopmental outcome. We believe that systematically comparison of model performance with and without biological sex as a confound to be a worthy endeavour. 

*Head circumference and birthweight
*Both head circumference and birthweight encode information about the overall health of neonates, and participants are welcome to also use these metadata as additional data for the prediction of birth age and neurodevelopmental outcome.  However, participants should be aware that:

1. These measures will not be available for all neonates
2. For neonates with available data, these are noisy measures (as is the case for any clinical data)
3. Both head circumference and birthweight should be interpreted in the context of gestational age at birth (preterm vs. term status), postmenstrual age at scan (age at measurement) and biological sex. Typically raw values are converted to centiles, using calculators such as those developed by INTERGROWTH21.

File Conversion¶

The cortical surface data released as part of the SLCN challenge are in gifti format (.shape.gii). However, in order to run SLCN on the Grand Challenge platform as a type II challenge, algorithms must load cortical metrics as .mha files rather than .shape.gii files. All images used to evaluate participant algorithms are .mha format, and have shape [40962,4].

After you have been granted access to the data, and are familiar with what it represents, the metrics need to be converted using simple Python code, which is located here.