Table 2.

ALSFRS slope prediction performance. RMSD and PCC are shown for all methods. FFNN, CNN, and RNN performance was obtained on the external test set (n = 731) using 10,000 bootstrap with resampling. For these predictors we reported the mean and the 95% confidence interval (CI). FFNN+CNN represents the ensemble prediction of the two neural networks. The best values for each metric are highlighted in bold. *Random Forest (RF) and Bayesian Additive Regression Trees (BART) are taken from literature³¹, where they are reported without CI.

Methods	RMSD	PCC
FFNN	0.528 (0.502–0.555)	0.451 (0.404–0.495)
CNN	0.527 (0.499–0.556)	0.439 (0.388–0.487)
RNN	0.529 (0.501–0.558)	0.429 (0.379–0.476)
FFNN+CNN	0.521 (0.494–0.548)	0.462 (0.415–0.508)
RF*	0.563	0.446
BART*	0.554	0.472