Table 3.
Per-Record QRS Detection Performance—MIT-BIH Arrhythmia Database.
| Record | Condition | SoC | BER | Sens. Base [%] | Sens. BAFE [%] | Spec. BAFE [%] | F1 BAFE [%] | ΔSens. [pp] | SQNR Base [dB] | SQNR BAFE [dB] | Clin. OK |
|---|---|---|---|---|---|---|---|---|---|---|---|
| MITBIH-001 | Normal Sinus Rhythm | 20% | 1.78 × 10⁻² | 37.5 | 97.8 | 97.7 | 97.8 | + 60.3 | 18.9 | 12.1 | Yes |
| MITBIH-001 | Normal Sinus Rhythm | 40% | 7.33 × 10⁻³ | 48.3 | 98.1 | 97.8 | 98.0 | + 49.8 | 22.7 | 13.8 | Yes |
| MITBIH-002 | Atrial Fibrillation | 20% | 1.78 × 10⁻² | 42.1 | 96.4 | 96.3 | 96.3 | + 54.3 | 18.6 | 11.6 | Yes |
| MITBIH-003 | Sinus Bradycardia | 20% | 1.78 × 10⁻² | 39.4 | 98.1 | 97.9 | 98.0 | + 58.7 | 20.0 | 11.5 | Yes |
| MITBIH-004 | Sinus Tachycardia | 20% | 1.78 × 10⁻² | 36.8 | 97.4 | 97.2 | 97.3 | + 60.6 | 18.6 | 11.6 | Yes |
| MITBIH-005 | PVC | 20% | 1.78 × 10⁻² | 31.2 | 72.3 | 81.6 | 76.7 | + 41.1 | 20.1 | 12.3 | No |
| MITBIH-006 | LBBB | 20% | 1.78 × 10⁻² | 28.9 | 96.2 | 96.1 | 96.2 | + 67.3 | 18.5 | 12.1 | Yes |
| MITBIH-007 | Atrial Flutter | 20% | 1.78 × 10⁻² | 23.8 | 95.6 | 96.0 | 95.8 | + 71.8 | 20.3 | 12.9 | Yes |
| MITBIH-008 | Normal Sinus Rhythm | 20% | 1.78 × 10⁻² | 14.1 | 97.7 | 97.9 | 97.8 | + 83.6 | 20.5 | 13.1 | Yes |
| MITBIH-009 | Ventricular Tachycardia | 20% | 1.78 × 10⁻² | 37.4 | 69.1 | 82.7 | 75.3 | + 31.7 | 18.6 | 12.5 | No |
| MITBIH-010 | Normal Sinus Rhythm | 20% | 1.78 × 10⁻² | 59.3 | 98.3 | 98.3 | 98.3 | + 39.0 | 18.9 | 11.8 | Yes |
| Mean (20% battery) | — | 20% | 1.78 × 10⁻² | 35.1 | 91.9 | 94.2 | 92.9 | + 56.8 | 19.3 | 12.1 | 8/10 |
| Mean (40% battery) | — | 40% | 7.33 × 10⁻³ | 43.0 | 91.8 | 93.5 | 92.6 | + 48.9 | 23.4 | 14.3 | 8/10 |
| Mean (100% battery) | — | 100% | 1.18 × 10⁻³ | 86.4 | 93.3 | 96.4 | 94.7 | + 7.0 | 31.4 | 17.2 | 8/10 |
ΔSens.: sensitivity gain (BAFE − baseline) in percentage points. Clin. OK: Se ≥ 95% (ANSI/AAMI EC5711). 8 of 10 MIT-BIH records meet this threshold at 20% SoC. ‡SQNR values represent 10 log₁₀(Pₛᴵᴳₙₐₗ/Pₙₒᴵₛᴵ). At ≤ 0% SoC, BAFE improves SQNR (+ 3.2 dB at 5% SoC) by suppressing severe bit-flip noise. At ≥ 20% SoC, BPF morphological distortion dominates: BAFE reduces SQNR by 7.2 dB—the approximate-computing trade-off33where wideband fidelity is sacrificed for QRS detectability.