Introduction
Materials and methods
Subjects
Non-SAHS | Mild SAHS | Moderate SAHS | Severe SAHS | |
---|---|---|---|---|
Number of patients | 2 | 12 | 5 | 5 |
Age (years) | 52.0 ± 15.6 | 48.6 ± 8.5 | 56.8 ± 6.4 | 46.6 ± 5.5 |
AHI (events/h) | 4.1 ± 5.7 | 9.9 ± 2.9 | 24.6 ± 3.9 | 43.8 ± 16.3 |
Epworth Sleepiness Score | 7.0 ± 8.5 | 11.6 ± 5.1 | 11.2 ± 6.9 | 12.4 ± 7.9 |
Study design
Signal preprocessing
Feature extraction
Index | Name | Definition |
---|---|---|
1 | Fmean, Fstd, Fran | Average, standard deviation, and range of tidal volume |
2 | Fha, Fhap | Number of breaths with a reduction more than 30% in tidal volume and its ratio to total number of breaths |
3 | Fap, Fapp | Number of breaths with a reduction more than 70% in tidal volume and its ratio to total number of breaths |
4 | Fnor, Fnorp | Number of breaths with a reduction less than 15% in tidal volume and its ratio to total number of breaths |
5 | Fkur | Fourth statistical moment in 0.2–0.4 Hz of NF’s frequency spectrum |
6 | Spstd, Spran | Standard deviation and range of SpO2 |
7 | Spten | Tendency of SpO2 |
8 | Spdum, Spdua | Duration of SpO2 desaturation |
9 | Spldm, Splda | Level of SpO2 desaturation |
10 | Sp92, Sp91 | Duration of SpO2 staying below 92 and 91% |
Design of cascading detector
Design of event detector
Results
Segment analysis
Reference | ACC (%) | SEN (%) | SPE (%) | |||
---|---|---|---|---|---|---|
AH | N | |||||
Estimated | AH | 33,429 | 46,247 | 88.3 | 75.2 | 89.6 |
N | 11,047 | 397,251 |
AHI analysis
Group | Number of AH events | Duration of AH events (min) | AHI (events/h) | |||
---|---|---|---|---|---|---|
Reference | Estimated | Reference | Estimated | Reference | Estimated | |
Non | 17.0 ± 7.0 | 9.0 ± 8.3 | 4.6 ± 2.3 | 6.2 ± 4.6 | 3.2 ± 1.4 | 1.7 ± 1.6 |
Mild | 61.5 ± 19.4 | 83.0 ± 37.2 | 18.8 ± 6.2 | 29.6 ± 7.9 | 9.9 ± 2.8 | 12.1 ± 4.2 |
Moderate | 132.2 ± 18.7 | 155.0 ± 23.6 | 42.5 ± 14.2 | 54.3 ± 15.5 | 24.6 ± 3.5 | 27.5 ± 5.4 |
Severe | 193.2 ± 63.2 | 200.2 ± 63.8 | 61.3 ± 21.2 | 72.2 ± 30.8 | 43.9 ± 14.5 | 45.3 ± 14.3 |
Determined from PSG | AHI cutoff (\) | |||||||||
---|---|---|---|---|---|---|---|---|---|---|
Non | Mild | Moderate | Severe | ≥ 5 | ≥ 15 | ≥ 30 | AVE | |||
Estimated | Non | 2 | 0 | 0 | 0 | SEN (%) | 100.0 | 100.0 | 100.0 | 100.0 |
Mild | 0 | 9 | 0 | 0 | SPE (%) | 100.0 | 78.6 | 94.7 | 91.1 | |
Moderate | 0 | 3 | 4 | 0 | PPV (%) | 100.0 | 76.9 | 83.3 | 86.7 | |
Severe | 0 | 0 | 1 | 5 | ACC (%) | 100.0 | 87.5 | 95.8 | 94.4 |
Discussion
Related work | Method | Signal | AHI cutoff | ACC (%) | SEN (%) | SPE (%) |
---|---|---|---|---|---|---|
Choi et al. [3] | Convolutional neural networks | Nasal pressure | 5 | 96.2 | 100.0 | 84.6 |
15 | 92.3 | 98.1 | 86.5 | |||
30 | 96.2 | 96.2 | 96.2 | |||
Gonzalo et al. [4] | AdaBoost-Linear discriminant analysis | Nasal flow | 5 | 86.5 | 87.1 | 80.0 |
15 | 81.0 | 85.9 | 72.9 | |||
30 | 82.5 | 74.2 | 90.6 | |||
Da Woon Jung et al. [7] | Rule-based | SpO2 | 5 | 97.8 | 98.6 | 94.4 |
10 | 96.7 | 98.4 | 92.9 | |||
15 | 95.7 | 96.4 | 94.6 | |||
30 | 96.7 | 97.1 | 96.5 | |||
Our study | Cascade of random forests | Nasal flow and SpO2 | 5 | 100.0 | 100.0 | 100.0 |
15 | 87.5 | 100.0 | 78.6 | |||
30 | 95.8 | 100.0 | 94.7 |