Skip to main content
[Preprint]. 2023 Sep 1:2023.06.19.545638. [Version 2] doi: 10.1101/2023.06.19.545638

Table 2:

Test performance of all models evaluated, displaying median R2 and median correlation of average hold-out performances (i.e. average across 5 outer folds) across 100 permutations.

Ensemble type Model name Median performance
Correlation R 2
None M1 CST-LL 0.395 0.132
Ipsi. SMATT-LL 0.441 0.175
L/R SMATT-LL 0.445 0.180
LBM 0.452 0.200
sLNM-LL 0.428 0.167
ChaCo (fs86) 0.458 0.204
ChaCo (shen268) 0.458 0.200
ChaCo (fs86) (feat. select.) 0.457 0.202
ChaCo (shen268) (feat. select.) 0.468 0.212
+ Demog. M1 CST-LL + demog. 0.424 0.173
Ipsi. SMATT-LL + demog. 0.457 0.199
L/R SMATT-LL + demog. 0.462 0.203
LBM + demog. 0.482 0.207
sLNM-LL + demog. 0.452 0.191
ChaCo (fs86) + demog 0.471 0.203
ChaCo (shen268) + demog 0.468 0.204
ChaCo (fs86) (feat. select.) + demog 0.471 0.205
ChaCo (shen268) (feat. select.) + demog 0.481 0.215
ChaCo M1 CST-LL + ChaCo (fs86subj) 0.461 0.210
Ipsi. SMATT-LL + ChaCo (fs86subj) 0.478 0.226
Ipsi. SMATT-LL + ChaCo (shen268) 0.481 0.228
L/R SMATT-LL + ChaCo (fs86subj) 0.481 0.227
L/R SMATT-LL + ChaCo (shen268) 0.481 0.228
LBM + ChaCo (fs86subj) 0.487 0.234
LBM + ChaCo (shen268) 0.491 0.240
sLNM-LL + ChaCo (fs86subj) 0.468 0.214
sLNM-LL + ChaCo (shen268) 0.472 0.217
M1 CST-LL + ChaCo (fs86subj) (feat. select.) 0.458 0.206
M1 CST-LL + ChaCo (shen268) (feat. select.) 0.469 0.217
Ipsi. SMATT-LL + ChaCo (fs86subj) (feat. select.) 0.476 0.223
Ipsi. SMATT-LL + ChaCo (shen268) (feat. select.) 0.485 0.233
L/R SMATT-LL + ChaCo (fs86subj) (feat. select.) 0.475 0.222
L/R SMATT-LL + ChaCo (shen268) (feat. select.) 0.486 0.232
LBM + ChaCo (fs86subj) (feat. select.) 0.487 0.235
LBM + ChaCo (shen268) (feat. select.) 0.495 0.243
sLNM-LL + ChaCo (fs86subj) (feat. select.) 0.462 0.208
sLNM-LL + ChaCo (shen268) (feat. select.) 0.474 0.222
ChaCo + Lesion load + Demog. M1 CST-LL + ChaCo (fs86subj) + demog. 0.476 0.214
M1 CST-LL + ChaCo (shen268) + demog. 0.477 0.216
Ipsi. SMATT-LL + ChaCo (fs86subj) + demog. 0.492 0.230
Ipsi. SMATT-LL + ChaCo (shen268) + demog. 0.495 0.232
L/R SMATT-LL + ChaCo (fs86subj) + demog. 0.494 0.231
L/R SMATT-LL + ChaCo (shen268) + demog. 0.494 0.232
LBM + ChaCo (fs86subj) + demog. 0.506 0.234
LBM + ChaCo (shen268) + demog. 0.509 0.237
sLNM-LL + ChaCo (fs86subj) + demog. 0.484 0.222
sLNM-LL + ChaCo (shen268) + demog. 0.487 0.224
M1 CST-LL + ChaCo (fs86subj) (feat. select.) + demog. 0.472 0.212
M1 CST-LL + ChaCo (shen268) (feat. select.) + demog. 0.482 0.222
Ipsi. SMATT-LL + ChaCo (fs86subj) (feat. select.) + demog. 0.490 0.228
Ipsi. SMATT-LL + ChaCo (shen268) (feat. select.) + demog. 0.498 0.237
L/R SMATT-LL + ChaCo (fs86subj) (feat. select.) + demog. 0.490 0.229
L/R SMATT-LL + ChaCo (shen268) (feat. select.) + demog. 0.499 0.237
LBM + ChaCo (fs86subj) (feat. select.) + demog. 0.505 0.234
LBM + ChaCo (shen268) (feat. select.) + demog. 0.513 0.241
sLNM-LL + ChaCo (fs86subj) (feat. select.) + demog. 0.479 0.218
sLNM-LL + ChaCo (shen268) (feat. select.) + demog. 0.492 0.230

Lighter blue shades represent better performance, and models are listed from theory-based to data-driven. Bemog. = demographics, Ipsi. = Ipsilesional, SMATT LL = sensorimotor tract template lesion load, L/R SMATT LL = left and right sensorimotor tract template lesion load, M1 CST LL = M1 corticospinal tract lesion load, ChaCo = Change in Connectivity, fs86 = FreeSurfer 86-region atlas, feat. select. = feature selection