Use of Novel Open-Source Deep Learning Platform for Quantification of Ki-67 in Neuroendocrine Tumors – Analytical Validation

Figures & data

Table 1 Descriptive Statistics for P1, P2, P3, P4, P5 and AI

Variables	Minimum	Maximum	Median	IQR*
P1**	4	88	25	37.1
P2	5	85	25	35
P3	4	80	30	40
P4	6	95	30	42
P5	1	80	30	42
AI	5	88	21	39

Notes: *Inter quartile range. **P1 counted the tumor cells manually and was considered as gold standard.

Table 2 Measurement of Agreement

		P1		Total	Kappa	Standard Error	Strength of Agreement
		≤ 20	> 20
P2	≤ 20	40	5	45	0.798	0.061	Good
	> 20	5	50	55
	Total	45	55	100
P3	≤ 20	35	6	41	0.674	0.074	Good
	> 20	10	49	59
	Total	45	55	100
P4	≤ 20	31	6	37	0.589	0.081	Moderate
	> 20	14	49	63
	Total	45	55	100
P5	≤ 20	31	5	36	0.609	0.079	Good
	> 20	14	50	64
	Total	45	55	100
AI	≤ 20	44	6	50	0.860	0.051	Excellent
	> 20	1	49	50
	Total	45	55	100

Figure 1 Digital image of Ki-67 neuroendocrine tumor prepared at 10x was uploaded in the software and the concordance between manual versus automated score was observed.

Table 3 Sensitivity, Specificity, PPV, NPV and Diagnostic Accuracy of P2, P3, P4, P5 & AI Using P1 as Gold Standard

		P1		Total	Sensitivity	Specificity	PPV*	NPV**	Diagnostic Accuracy
		≤ 20	> 20
P2	≤ 20	40	5	45	88.89%	90.91%	88.89%	90.91%	90%
	> 20	5	50	55
	Total	45	55	100
P3	≤ 20	35	6	41	77.78%	89.09%	85.37%	83.05%	84%
	> 20	10	49	59
	Total	45	55	100
P4	≤ 20	31	6	37	68.89%	89.09%	83.78%	77.78%	80%
	> 20	14	49	63
	Total	45	55	100
P5	≤ 20	31	5	36	68.89%	90.91%	86.11%	78.12%	81%
	> 20	14	50	64
	Total	45	55	100
AI	≤ 20	44	6	50	97.78%	89.09%	88%	98%	93%
	> 20	1	49	50
	Total	45	55	100

Notes: *Positive predictive value. **Negative predictive value.

Figure 2 Validity by Correlation Matrix showing that the finding of Ki-67 by P1 is strongly positively correlated with P2, P3 and P4 while very highly positively correlated with P5 and AI.

Figure 3 ROC Plots prepared for pathologists P2, P3, P4, P5 and AI to access the quality of test by taking Pathologist P1 score as gold standard.