A validated natural language processing algorithm for brain imaging phenotypes from radiology reports in UK electronic health records

Table 1 Dataset statistics: number of reports, sentences, entity, modifier and phenotype (label) annotation per data set (ESS dev/test vs Tayside dev/test) for annotator 1

	ESS Dev	ESS Test	Tayside Dev*	Tayside Test*
Reports	364	266	362	700
Sentences	3837	2855	2791	3948
Tokens	32,229	22,842	50,522	48,519
Total Entities	4332	2924	2997	2986
Disease Entities	2373	1494	1361	1501
Modifier Entities	1959	1430	1636	1485
Total Phenotypes (Labels)	792	518	558	506
Atrophy	187	122	90	164
Small vessel disease	245	159	60	145
Stroke, underspecified	24	15	16	< 5
Haemorrhagic stroke, deep, old	2	4	< 5	< 5
Haemorrhagic stroke, deep, recent	2	2	< 5	< 5
Haemorrhagic stroke, lobar, old	4	3	7	< 5
Haemorrhagic stroke, lobar, recent	1	4	< 5	< 5
Haemorrhagic stroke, underspecified	7	10	94	15
Ischaemic stroke, cortical, old	112	61	27	26
Ischaemic stroke, cortical, recent	21	14	19	12
Ischaemic stroke, deep, old	140	85	60	41
Ischaemic stroke, deep, recent	7	4	< 5	< 5
Ischaemic stroke, underspecified	5	12	85	15
Haemorrhagic transformation	1	1	10	< 5
Subdural haematoma	9	6	20	8
Subarachnoid haemorrhage, aneurysmal	1	0	< 5	< 5
Subarachnoid haemorrhage, other	6	6	21	7
Microbleed, deep	2	1	< 5	< 5
Microbleed, lobar	2	1	< 5	< 5
Microbleed, underspecified	0	1	< 5	< 5
Tumour, glioma	0	0	< 5	< 5
Tumour, meningioma	2	4	< 5	< 5
Tumour, metastasis	2	0	22	37
Tumour, other	10	3	21	12

*Small numbers suppressed in the NHS Tayside table due to data governance requirements