From: Deep learning based automatic segmentation of organs-at-risk for 0.35 T MRgRT of lung tumors
Function | Parameter | Tested range | Final value | |
---|---|---|---|---|
Phase 1 | Phase 2 | |||
Probability | \(p_{\rm{aug}}\) | 0.25–1 | \(p_1=0.6\) | \(p_2=0.85\) |
Learning rate | lr | \(10^{-5}\)–\(2\times 10^{-2}\) | \(10^{-3}\) | |
Spatial | ||||
Rotation | \(\alpha _{\rm{max}}\,[^\circ ]\) | 5–20 | 15 | |
Translation | \(\Delta _{\rm{max}}\,[{\hbox {mm}}]\) | \(15-30\) | 22.5 | |
Zooming | \(z_{\rm{min}},\, z_{\rm{max}}\) | – | \(0.9,\,1.1\) | |
Deformation | \(n_{\rm{cp}}\) | 5–20 | – | 8 |
\(d \,[{\hbox {mm}} ]\) | 15–45 | – | 24 | |
MR | ||||
Motion | \(m_{{\alpha }}[^\circ ]\) | 0–15 | – | 10 |
\(m_{\Delta }\,[{\hbox {mm}}]\) | 15–75 | – | 45 | |
Bias field | order | 1–3 | – | 1 |
\(c_{\rm{mag}}\) | 0–1 | – | 0.4 | |
Noise | \(\sigma\) | 0.01–0.25 | 0.05 | 0.1 |
\(\mu\) | – | 0 |