Table 3 Highest predictive models and best performing feature sets.

From: Predicting 2-year neurodevelopmental outcomes in preterm infants using multimodal structural brain magnetic resonance imaging with local connectivity

BSID-III

Group

Top prediction model (predictor sets)

RMSE

Variance explained (%)

Cognitive

Preterm

ElasticNet (Feature set A, C, and D)

13.352

17

EP

RF (Feature set C and D)

15.402

13

V-LP

ElasticNet (Feature set A, B, and D)

11.205

17

Motor

Preterm

ElasticNet (Feature set C)

12.996

13

EP

XGBoost (Feature set A, C, and D)

11.363

15

V-LP

RF (Feature set A and B)

13.698

10

Language

Preterm

XGBoost (Feature set A)

11.792

15

EP

XGBoost (Feature set B and C)

11.674

3

V-LP

ElasticNet (Feature set A, C, and D)

12.425

6

  1. Feature set A: Local connectivity features; B: Clinical characteristic features; C: Volumetric features; D: Global connectivity features.
  2. EP, extremely preterm; V-LP, very-to-late preterm; RF, random forest; BSID-III, Bayley Scales of Infant and Toddler Development, Third Edition; RMSE, root mean squared error.
Back to article page