Skip to main content

Table 1 Summary of previous studies of automatic detection of radiology reports with actionable findings, along with this study

From: Automatic detection of actionable radiology reports using bidirectional encoder representations from transformers

 

Target language

Multiple diseases

Use of labels in clinical practice

Criteria for positive class

Target sections in radiology reports

Methods

Meng et al. [7]

English

Yes

No

Expressions suggesting the need to promptly communicate to the referring clinician

Impression

Existing tool

Helibrun et al. [8]

English

Yes

No

Expressions suggesting specific critical findings

Impression

Existing tool

Carrodeguas et al. [9]

English

Yes

No

Follow-up recommendations

Impression

SML, LSTM

Yetisgen-Yildiz et al. [10]

English

Yes

No

Follow-up recommendations

Order information, findings, impression

SML

Yetisgen-Yildiz et al. [11]

English

Yes

No

Follow-up recommendations

Order information, findings, impression

SML

Dutta et al. [12]

English

Yes

No

Follow-up recommendations

Findings, impression, recommendation

Existing tool

Lau et al. [13]

English

Yes

No

Follow-up recommendations

(Not specified)

GRU

Dang et al. [14]

English

Yes

No

Follow-up recommendations

(Not specified)

Decision tree

Imai et al. [15]

Japanese

Yes

No

Expressions suggesting malignancy

Findings

Syntactic analysis

Lou et al. [16]

English

No

Yes

Reports pointing at indeterminate or suspicious upper abdominal mass

(Not specified)

SML

Danforth et al. [17]

English

No

Yes

ICD-9 codes suggesting lung nodules

(Not specified)

Rule base

Garla et al. [18]

English

No

Yes

Expressions suggesting potentially malignant liver lesions

(Not specified)

SML

Farjah et al. [19]

English

No

No

Expressions suggesting lung nodules

(Not specified)

Existing tool

Gershanik et al. [20]

English

No

No

Expressions suggesting lung nodules

Findings, impression

Existing tool

Oliveira et al. [21]

English

No

No

Expressions suggesting incidental lung nodules

Order information, findings

Rule base

Pham et al. [22]

French

No

No

Expressions suggesting incidentalomas

Order information, findings, impression

SML

Mabotuwana et al. [23]

English

No

No

Follow-up recommendations

(Not specified)

Rule base

Morioka et al. [24]

English

No

No

Expressions suggesting abdominal aorta aneurysm

(Not specified)

Existing tool

Xu et al. [25]

English

(Not specified)

No

Follow-up recommendations

Order information, findings, impression

SML

Fu et al. [26]

English

No

No

Expressions suggesting silent brain infarction or white matter disease

(Not specified)

Rule base, SML, CNN

This study

Japanese

Yes

Yes

Reports with an actionable tag

Order information, findings, impression

SML, LSTM, BERT

  1. BERT = Bidirectional Encoder Representations from Transformers, CNN = Convolutional Neural Network, GRU = Gated Recurrent Units, LSTM = Long Short-Term Memory, and SML = Statistical Machine Learning