Table 3 Classification and segmentation metrics of the different models tested with and without augmentation in training, validation and test sets.

Classification		Train					Validation					Test
		Acc	Pr	Rc	F1	Auc	Acc	Pr	Rc	F1	Auc	Acc	Pr	Rc	F1	Auc
ResNet		93	93	93	93	93	74	74	75	73	75	61 [57–66]	71 [64–78]	55 [53–57]	47 [44–52]	55
ResNet	*aug	88	88	88	88	88	77	77	76	78	76	75 [71–79]	76 [72–80]	73 [69–76]	73 [69–77]	73
EfficientNet		92	92	92	92	92	75	74	76	74	76	58 [54–63]	58 [48–66]	52 [50–54]	43 [40–47]	52
EfficientNet	*aug	87	87	87	87	87	76	77	78	76	78	79 [75–83]	78 [75–82]	79 [75–82]	79 [75–82]	79
DenseNet		93	93	93	93	93	76	75	77	75	77	57 [53–62]	29 [27–31]	50 [50–50]	36 [35–38]	50
DenseNet	*aug	89	89	89	89	89	78	77	79	77	79	73 [69–77]	75 [72–80]	70 [66–74]	70 [66–74]	70
Swin-T		86	86	86	86	86	81	80	82	80	82	64 [60–68]	69 [63–74]	59 [56–62]	55 [51–60]	59
Swin-T	*aug	80	80	80	80	80	81	81	81	81	81	78 [75–82]	78 [75–82]	79 [75–82]	78 [75–82]	79
UNI2		85	85	85	85	85	80	80	80	80	80	66 [62–70]	74 [70–77]	69 [66–73]	65 [61–69]	69
UNI2	*aug	79	79	79	79	79	79	79	80	79	80	63 [59–67]	74 [71–77]	68 [64–70]	61 [57–66]	67
Segmentation
		Dice	IoU				Dice	IoU				Dice	IoU
Unet		21	13				18	11				16	12
Unet	*aug	14	8				14	9				26	22
DeepLabV3		57	41				33	23				24	15
DeepLabV3	*aug	29	18				23	14				24	14
SegFormer		63	49				52	44				46	37
SegFormer	*aug	31	20				38	30				25	17
Glomeruli (SegFormer)		97 ± 1	92 ± 1				95 ± 1	89 ± 1				94	88
Glomeruli (SegFormer)	*aug	95 ± 1	89 ± 1				94 ± 1	88 ± 1				94	88

Quick links

Search