Table 1.
Methods of assessing auto-contouring using deep learning contouring (DLC) for prostate structures.
| Centre 1 | Centre 2 | Centre 3 | |
|---|---|---|---|
| Number of patients for timing | 9 | 42 manual, 42 DLC | 10 |
| Number of patients for quantitative analysis | 9 | 10 | N/A |
| Same patients for timing and quantitative analysis? | Y | N | N/A |
| Timing Study design | Paired | Unpaired | Paired |
| Number of staff | 2 (both outlined 9 each) | 4 | 10 for manual, 1 for DLCExpert |
| Staff group | Radiotherapy Planners | Radiotherapy Planners | Radiotherapy Planners and clinicians. One clinician edited DLCExpert contours |
| Existing method | Manual | Manual | Manual |
| CT scanner/slice thickness | Philips Brilliance Big Bore (3 mm) | GE Discovery (2.5 mm) | GE Discovery (2.5 mm) |
| DLCExpert model name | Prostate_CT_NL006_MO | Prostate_CT_NL005_GN (bladder and rectum)Prostate_CT_NL010_NN (femoral heads) | Prostate_CT_NL006_MO |
| DLCExpert model description | Generic model based on data from a centre in the Netherlands. | Generic models based on data from a centre in the Netherlands. | Generic model based on data from a centre in the Netherlands. |
| Number of images for training model | 437 | 242 (bladder and rectum)337 (femoral heads) | 437 |
| OARs | Bladder, femoral heads, rectum | Bladder, femoral heads, rectum | Bladder, femoral heads, rectum |
| Targets | None | None | Prostate, seminal vesicles |
| Editing software | Pinnacle v16 (Philips, NL) | RayStation v7 (RaySearch, Sweden) | Eclipse v13.7 (Varian Medical Systems, Palo Alto, USA) |
| Outlining protocol basis | CHHiP | CHHiP, RTOG atlases | CHHiP |
| Timing method | Manual- time to draw or edit timed by the staff member | Manual- time to draw or edit timed by the staff member | Manual- time to draw or edit timed by the staff member Aria (Varian Medical Systems, Palo Alto, USA)-time to draw CTVs |