Table 6.

Results for celiac disease baseline and KD approaches as percentages with corresponding 95% CIs. Baseline models were trained from scratch until convergence on the corresponding magnification. The KD model without an auxiliary dataset was trained using the labeled dataset. Boldface text indicates the best-performing model for each magnification and metric.

	Celiac Disease
	Baseline	KD	KD (ADv1)	KD (ADv2)
mag = 0.625x (16 μm/pixel)
Accuracy	79.11 (77.74–80.54)	81.31 (79.91–82.72)	82.55 (81.15–83.97)	83.17 (81.75–84.61)
F1-Score	55.72 (52.08–59.34)	64.16 (60.92–67.37)	64.95 (61.43–68.45)	66.83 (63.46–70.14)
Precision	56.20 (52.40–59.96)	64.27 (60.91–67.56)	66.65 (62.94–70.32)	67.21 (63.78–70.52)
Recall	55.67 (52.17–59.16)	65.13 (62.06–68.20)	64.11 (60.64–67.61)	69.29 (66.06–72.42)
mag = 1.25x (8 μm/pixel)
Accuracy	82.70 (81.23–84.14)	84.03 (82.61–85.47)	84.43 (83.01–85.85)	84.87 (83.40–86.32)
F1-Score	65.06 (61.55–68.43)	70.49 (67.45–73.55)	69.75 (66.53–72.94)	71.20 (67.89–74.40)
Precision	65.06 (61.51–68.48)	70.53 (67.39–73.66)	69.32 (66.13–72.51	71.14 (67.81–74.35)
Recall	65.22 (61.70–68.63)	71.06 (68.10–74.02)	70.95 (67.72–74.17)	73.56 (70.40–76.61)
mag = 2.5x (4 μm/pixel)
Accuracy	83.71 (82.29–85.17)	85.68 (84.25–87.13)	85.38 (83.94–86.78)	85.83 (84.38–87.27)
F1-Score	68.32 (64.92–71.66)	73.01 (69.94–76.03)	72.39 (69.21–75.41)	73.56 (70.42–76.64)
Precision	68.23 (64.77–71.67)	74.74 (71.57–77.90)	72.99 (69.76–76.06)	73.61 (70.44–76.68)
Recall	68.57 (65.13–71.98)	74.67 (71.99–77.28)	75.67 (72.86–78.34)	76.43 (73.61–79.17)
mag = 5x (2 μm/pixel)
Accuracy	86.15 (84.71–87.61)	85.74 (84.28–87.21)	86.99 (85.54–88.46)	87.20 (85.78–88.62)
F1-Score	73.42 (70.15–76.63)	73.27 (70.19–76.33)	75.07 (71.89–78.17)	75.86 (72.71–78.92)
Precision	73.44 (70.12–76.68)	75.10 (71.91–78.23)	76.46 (73.42–79.44)	76.07 (72.95–79.13)
Recall	73.65 (70.41–76.93)	74.82 (72.16–77.51)	78.00 (75.18–80.72)	77.41 (74.41–80.35)