Table 5 Tractography cross-validation results on the TractoInferno dataset. The Prob-vMF model did not produce valid results, and is noted as {N/A}.

From: TractoInferno - A large-scale, open-source, multi-site database for machine learning dMRI tractography

 

Dice

Overlap

Overreach

Reference methods

Deterministic

0.409 ± 0.031

0.276 ± 0.027

0.027 ± 0.005

Probabilistic

0.571 ± 0.032

0.448 ± 0.031

0.066 ± 0.009

PFT

0.691 ± 0.078

0.692 ± 0.120

0.257 ± 0.062

SET

0.638 ± 0.044

0.590 ± 0.072

0.188 ± 0.047

Ensemble (Det + Prob + PFT + SET)

1.000 ± 0.000

1.000 ± 0.000

0.000 ± 0.000

RNN-based methods

Det-SE (Learn2track)

0.610 ± 0.022

0.539 ± 0.032

0.206 ± 0.039

Det-Cosine

0.599 ± 0.043

0.528 ± 0.052

0.198 ± 0.030

Prob-Sphere (DeepTract)

0.550 ± 0.051

0.449 ± 0.052

0.128 ± 0.027

Prob-vMF (Entrack)

N/A

N/A

N/A

Prob-Gaussian

0.561 ± 0.049

0.475 ± 0.059

0.166 ± 0.041

Prob-Mixture

0.489 ± 0.056

0.381 ± 0.059

0.101 ± 0.033

Back to article page