Deep learning-based segmentation of multisite disease in ovarian cancer

European Radiology Experimental

Table 1 Composition of the three datasets (total number of scans = 451) including information on voxel spacing and disease expression along all available time points

Dataset	Training	Validation	Test
Number of scans	276	104	71
Pretreatment scans	157	53	71
Post-NACT scans	119	51	0
Patient age [years]
Median	65.5	65.5	63
Min–max	29–90	35–85	41–80
Pixel spacing [mm]
Median	0.68	0.76	0.77
Min–max	0.53–0.93	0.61–0.96	0.57–0.98
Slice thickness [mm]
Median	5.0	5.0	5.0
Min–max	1.25–5.0	1.5–5.0	2.0–7.5
Pelvic/ovarian tumour
Number of scans showing tumour in this location	246	102	69
Mean volume [cm³]	275	241	381
Mean number of connected components	2.4	2.6	2.4
Omental tumour
Number of scans showing tumour in this location	198	98	56
Mean volume [cm³]	119	197	146
Mean number of connected components	6.7	5.3	5.7

Due to the anonymisation of the datasets, no information on the ethnicity of the patients was prevalent and the patient age was not available in 14 out of 71 patients in the test data. Additional information on the acquisition settings of the computed tomography examination is given in the Supplementary Materials. NACT Neoadjuvant chemotherapy