Fig. 2: The relationship between prediction accuracy and total scan duration (sample size × scan time per participant).
From: Longer scans boost prediction and cut costs in brain-wide association studies
a, The prediction accuracy (Pearson’s correlation) of the cognitive factor score as a function of the total scan duration (defined as the number of training participants × scan time per participant). There are 90 dots in the ABCD plot (left) and 174 dots in the HCP plot (right). Each colour shade represents a different total number of participants used to train the prediction algorithm. The asterisk indicates that all available participants were used and the sample size is therefore close to, but not exactly, the number shown. In both datasets, there were diminishing returns of both sample size and scan time, whereby each unit increase in sample size or scan duration resulted in progressively smaller gains in prediction accuracy. In the HCP dataset, the diminishing returns of scan time were more prominent beyond 30 min (Supplementary Table 1). b, The normalized (norm.) prediction accuracy of the two cognitive factor scores and 34 other phenotypes versus log2[total scan duration], ignoring data beyond 20 min of scan time. Cognitive, mental health, personality, physicality, emotional and well-being measures are shown in shades of red, grey, blue, yellow, green and pink, respectively. The black line shows that the logarithm of total scan duration explained prediction performance well across phenotypic domains and datasets. The Pearson’s correlation was computed between the log of total scan duration and normalized prediction performance based on 2,520 dots in the panel (16 total scan durations × 90 ABCD phenotypes + 18 total scan durations × 60 HCP phenotypes = 2,520). P values were computed using subsampling (to ensure independence) and 1,000 permutations (Supplementary Table 1). Attn prob, attention problems; cog, cognition; cryst, crystalized; disc, discounting; emo match, emotional face matching; ep mem, episodic memory; exec funct, executive function; flex, flexibility; int, intelligence; mem, memory; orient, orientation; proc spd, processing speed; PSQI, Pittsburgh Sleep Quality Index; rel proc, relational processing; satis, satisfaction; SusAttn (spec), sustained attention (specificity); vocab, vocabulary; vs, visuospatial.
