APR31a has been developed to perform better than APR30a. APR31a performance is characterized by:
- its accuracy in positioning systolic and diastolic points,
- its ability to perform better than APR30a when the latter produced unsatisfactory results (e.g. absent or erroneous detection).
These characteristics were evaluated on a validation database (49 files) including arterial blood pressure recordings, from various species and experimental conditions, on which APR30a detection failed. Systolic and diastolic points were manually annotated by NOTOCORD experts. Lower and upper acceptance margins for timings and amplitudes were defined for each set of points (systolic and diastolic separately) according to manual positioning.
APR31a performance was assessed by comparing manual and automated analysis, using the following criteria:
- Detection sensitivity: Se,
- Detection positive predictivity: P+.
- Computation details of Se and P+ are given at the end of the page.
APR31a was validated when achieving higher mean detection predictivity than APR30a on the same database.
The overall result of detection performance (Se and P+) for systolic and diastolic pressure position for all recordings is given in Table below, presented as “mean [min/max]” values.
Differences with APR30a
Average P+ and Se for systolic and diastolic detections are higher with APR31a than with APR30a. Shown below the detection improvement on a file processed by both APR31a and APR30a. Systolic and diastolic pressures are now correctly detected with APR31a.
Improved detection with APR31a (vs APR30a) on a blood pressure signal recording (File #12).
Cross: APR31a systolic (red) and diastolic (blue) pressures; Bar: APR30a systolic (orange) and diastolic (light blue).
An Analysis Error marker is generated when APR31a cannot detect systolic and/or diastolic pressure, or in the presence of noise artifacts such as saturation.
Blood pressure signal recording (File #29) with a saturation event. APR31a did not analyze the artifact and produced an Analysis Error marker.
Cross: APR31a systolic (red) and diastolic (blue) pressures; Bar; APR30a systolic (orange) and diastolic (light blue).
Average detection predictivity for diastolic pressure is above 99%, markedly increased compared to APR30a. On File #25, APR31a showed a slightly lower predictivity than APR30a, which can be explained by changes in baseline and cycle morphology. Such changes lead to uncertainty and lower confidence even with the manual positioning of diastolic pressure.
Blood pressure signal recording (File #25) on which end-diastole position is difficult to identify.
Cross: APR31a systolic (red) and diastolic (blue) pressures; Bar; Manual positioning of systolic (orange) and diastolic (light blue) pressures.
Computation of performance criteria
Number of true positives in a file. A true positive occurs when automated analyzer detects a point within the acceptance margins around the manually annotated point.
Number of false positives in a file. A false positive occurs when automated analyzer detects a point, but not located within the acceptance margins around the manually annotated point.
Number of false negatives in a file. A false negative occurs when automated analyzer fails to detect a point annotated manually.
Detection sensitivity measures the analyzer ability to detect every fiducial point. Se is computed using the following formula:
Detection positive predictivity measures the analyzer ability to detect fiducial points only. P+ is computed using the following formula: