Table 3.

ML device output by type

Output type	Devices	Description
Quantification	13	Quantification of information derived from the images, such as, cardiac function and blood flow,23 30 32 33 37 or volume of structures including the brain,31 35 36 70 and prostate.34
Triage notifications	10	Triage notification alert clinicians to cases with suspected positive findings.25 40–48
Case-level finding of disease	6	Case level finding of disease such as, wrist fractures,52 diabetic retinopathy,49 osteoarthritis,51 heart murmurs24 50 and cardiac arrhythmias.24 53
Identify features of disease	6	Identify features of disease thereby drawing clinician attention to them, such prompting breast55–57 59 or lung58 cancers on images or cardiac arrythmias on ECG tracings.54
Clinical grading or scoring	5	Clinical grading or scoring (n=5) on standardised clinical assessment instruments, such as BI-RADS,60 63 64 LI-RADS,62 lung-RADS,62 or Agatston-equivalent scores.61
Enhanced images	4	Enhanced images with reduced noise and improved image quality.26–29
Automatic coding of features or events	2	Automatic coding of features or events in the data, such as sleep stages and respiratory events in polysomnography data,39 or colour coding structures in optical coherence tomography.38
Automatic control of electronic or mechanical devices	2	Automatic control of electronic or mechanical devices, such as fluoroscope collimator67 and automatic recording of ultrasound clips dependent on detected image quality.66
Treatment recommendations	1	Treatment recommendations, such as adjustments to insulin pump dose ratios.87

BI-RADS, Breast Imaging-Reporting and Data System; LI-RADS, Liver Imaging-Reporting and Data System; ML, machine learning.