Automated segmentation of lesions and organs at risk on [68Ga]Ga-PSMA-11 PET/CT images using self-supervised learning with Swin UNETR

Table 2 The averaged precision metric results per-patient in five distinct experiments across all targets on the test set

	Precision
Approach Target	nnU-Net – PET	nnU-Net -PET/CT	Proposed method - PET	Proposed method - PET/CT (without SSL pre-training)	Proposed method - PET/CT (SSL pre-training + fine-tuning)
Lacrimal glands	0.63	0.65	0.65	0.73	0.76
Parotid glands	0.80	0.84	0.82	0.84	0.89
Tubarial glands	0.63	0.66	0.64	0.64	0.74
Sublingual glands	0.55	0.61	0.66	0.64	0.70
Submandibular glands	0.80	0.80	0.80	0.79	0.84
Spleen	0.79	0.83	0.81	0.86	0.86
Liver	0.86	0.88	0.88	0.90	0.92
Bowel	0.69	0.73	0.68	0.72	0.78
Kidneys	0.74	0.80	0.76	0.83	0.84
Bladder	0.72	0.75	0.73	0.77	0.81
Lesions	0.52	0.58	0.54	0.60	0.65
Mean	0.70	0.74	0.72	0.75	0.80

ISSN: 1470-7330