Table 2.

Comparison of advanced DL image analysis studies in digital pathology, comprising mutation prediction, prognostication and response prediction.

Reference	Description	Ext. validation	Number of slides	Number of patients	Number of cohorts	AUROC	F score	Accuracy	Other metrics
Mutation detection
69	Prediction of ER status in breast cancer	No	859	859	1	N/A	N/A	84%	Sensitivity = 88%; specificity = 76%
36	Prediction of SPOP mutation in prostate cancer	Yes	365	N/A	2	0.86 (P = 0.0038)	N/A	N/A	N/A
26	Prediction of different genes in lung cancer and ext. validation of EGFR mutation	Yes	1975	N/A	3	0.68	N/A	N/A	N/A
37	Prediction of BRAF and NRAS in melanoma	Yes	361	N/A	2	0.75 (BRAF); 0.77 (NRAS)	N/A	N/A	N/A
70	TP53 mutation prediction	No	27,815	N/A	28	0.8 (stomach)	N/A	N/A	N/A
71	Detection of HPV in head and neck cancer; detection of EBV in gastric cancer	Yes	1031	1031	4	0.7 (HPV); 0.81 (EBV)	N/A	N/A	N/A
27	Prediction of microsatellite instability in colorectal, gastric and endometrial cancer	Yes	2108	1952	5	0.84 (CRC)	N/A	N/A	N/A
38	Pan-cancer prediction of gene expression	No	10,514	8725	28	0.81 (MSI)	N/A	N/A	N/A
72	Prediction of BAP1 expression in uveal melanoma	Yes	47	47	2	N/A	0.93	92.8%	Sensitivity = 92.1%; specificity = 91.1%
73	Prediction of tumour mutational burden in liver cancer	No	368	350	1	0.95	N/A	94.86%	N/A
74	Prediction of PD-L1 status in non-small- cell lung cancer patients	No	130	130	N/A	0.8 (P < 0.01)	N/A	N/A	N/A
Therapy-response prediction
51	Prediction of response to ipilimumab in melanoma patients	No	31	31	1	N/A	N/A	70.9%	N/A
52	Prediction of probability that tissue from non-small-cell lung cancer will respond to immunotherapy	No	56	56	2	0.65	N/A	N/A	N/A
Survival prediction
						C score	Hazard ratio
48	Prediction of 5-year disease-specific survival in patients with colorectal cancer	No	420	420	1	N/A	2.3		AUROC = 0.96
75	Consensus molecular subtyping of colorectal cancer and predication of overall survival	No	769	N/A	2	0.8	N/A		N/A
45	Prediction of survival in colorectal cancer	Yes	1382	N/A	3	N/A	1.63 (1.14–2.33, P = 0.008)		N/A
47	Prediction of survival for patients with intrahepatic cholangiocarcinoma	No	246	246	2	N/A	0.86		N/A
76	Classification of patients to high risk or low risk in order to predict overall survival	No	1299	1299	2	N/A	1.74 (1.16–2.61, P = 0.006)		AUROC = 0.58
77	Stratification of patients into groups of short- and long-term survival by means of tumour- infiltrating lymphocytes	No	70	70	1	0.87	N/A		N/A
46	Prediction of survival in mesothelioma and identification of histological correlates	Yes	3037	3037	2	0.66	N/A		N/A
49	Prediction of development of metastatic recurrence in primary melanoma patients	Yes	263	263	5	N/A	N/A		AUROC = 0.91
78	Stratification of patients with colorectal cancer to good, uncertain or poor prognosis	Yes	4515	3595	4	N/A	3.83		Accuracy = 76%; sensitivity = 52%; specificity = 78%; PPV = 0.19; NPV = 0.94
79	Classification of patients with brain cancer in four groups based on survival time after diagnosis	Yes	664	454	2	N/A	N/A		AUROC = 0.96; accuracy = 80%
80	Prediction of overall survival of patients with hepatocellular carcinoma	Yes	732	522	2	0.7	4.3		N/A
81	Prediction of disease-specific survival in ten different cancer types	No	12,095	4880	10	61.1 (57.2, 65.1)	1.48 (P < 0.0001)		AUROC = 0.64 (58,70.3); 5-year disease- specific survival

N/A not available.

For each study, the level of evidence (presence of external validation), the number (#) of tissue slides, patients and patient cohorts as well as quantitative performance metrics are given, including area under the receiver-operating curve, AUROC; F score; accuracy; positive predictive value, PPV; negative predictive value, NPV; true-positive rate, TPR; false-positive rate, FPR; false-negative rate, FNR; other metrics (sensitivity, specificity and others) if reported in the study. This table is related to Fig. 2b.