Objective: Detecting heartbeats in photoplethysmography (PPG) underlies heart-rate, heart-rate-variability and arrhythmia analysis on wearable and clinical devices, and must be both efficient and accurate. Several accurate benchmark detectors, especially the multi-scale peak-and-trough family (MSPTD/MSPTDfast), rely on costly local-maxima scalograms; we seek a detector that improves the benchmark-wide speed–accuracy trade-off rather than merely accelerating one baseline. Approach: MorphSSF converts each PPG window to a slope-sum function (SSF), estimates the dominant pulse period from a band-limited autocorrelation, and selects candidates as fixed points of a period-matched running-maximum dilation on the SSF. This combines systolic-upstroke enhancement with adaptive period spacing using one-dimensional operations and no scalogram. We evaluate under identical conditions on seven datasets (544 recordings) with benchmark-wide rank tests, detector-specific exclusions and one-to-one matching. Separate fixed-point C and causal MATLAB variants are evaluated. Main results: MorphSSF has the best mean dataset F1 rank. Against MSPTDfast (v2), sensitivity tests favour MorphSSF on five datasets (including a zero median effect on CapnoBase); F1 is higher on MIMIC-test, MIMIC-train and PPG-DaLiA, while positive predictive value (PPV) is lower on WESAD. Both pooled and equal-dataset analyses favour MorphSSF in F1 against all eight baselines. MorphSSF is 3.05–218.18× faster than the eight baselines, including 3.23× faster than MSPTDfast (v2), based on median runtime ratios across recordings. All detectors receive the same prefiltered PPG input, with this shared filtering step excluded from runtime measurements. Significance: Period-adaptive detection on the slope-sum provides an efficient, scalogram-free approach to PPG beat detection.
Objective Electrical Impedance Tomography (EIT) is a developing medical imaging device. It is a non-invasive modality that operates by injecting a small amount of current into the body and measuring the boundary voltages. An ill-posed inverse problem is solved to determine an estimate of the internal impedance of the body. Due to the ill-posed nature and its associated physics, inclusions with different geometries may appear similar in reconstructed images. This work aims to identify differences between multiple inclusions of varying geometries utilizing a post processing method. It is hypothesized that the spatial variance of the reconstructed inclusion should change proportionately to the central diameter of the inclusion and the peak relative change in impedance should be consistent with the material type of the inclusion. Approach Several inclusions with different vertical geometries were imaged using a validate open-source EIT device at different positions in a saline phantom. The reconstructed images were analysed by considering a 1D slice of the impedance values. This 1D slice was approximated by a Gaussian function. It is proposed that the Gaussian parameters associated with an inclusion should change based on the inclusion geometry and material. Main Results When the 1D slice was coincident with the inclusion the Gaussian function was a good approximation (R2 ≥ 0.93). The Gaussian parameters followed repeatable patterns as the inclusion was moved to different positions. The amplitude was shown to only change with respect to the inclusion’s volume and not the geometry, as the image slice is effectively the average of the electric field solution. The standard deviation was consistent at 10 mm diameter irrespective of the inclusion geometry. Significance This simplified method was unable to distinguish between vertical inclusion geometries compared to the full EIT solution. However, the position and size of the inclusions can be recovered from the 1D slices. Thus, the same information about the inclusion can be determined from a 1D set of impedances as a 2D image.
Abstract Objective . Wearable single-lead ECG has expanded arrhythmia screening, but many recordings are inconclusive, and each corpus defines that word differently. We ask how much deployment error is un-analyzable input rather than misclassification when analyzability is measured with one instrument. Approach . We evaluate three diagnostic databases in-domain and leave-dataset-out under a strict loss counting windows carrying each corpus’s inconclusive label as errors unless deferred. We score analyzability with an unsupervised morphological criterion calibrated against three experts on free-living ECG. Reference unreadability is sparse and imbalanced: only seven of 15 subjects contain an unreadable window and one supplies 88% of all unreadable windows. Thresholded sensitivity and every prevalence or true-share correction are therefore exploratory. Main results . The criterion flags almost all expert-unreadable windows but only subsets of the windows labeled inconclusive by their corpora: an inconclusive rate measures convention as much as signal. At τ a = 0.50 , criterion-positive windows are 8.1%, 56.7% and 69.6% of the strict proxy loss on Holter, handheld and patch. The true-share point intervals lie below 50% on handheld and above it on patch, but joint calibration-and-sampling envelopes span 50% on both and still do at the Youden threshold 0.60. On handheld, the calibration-only envelope is more than twice as wide as the corpus-sampling-only envelope. Gate-versus-confidence ordering changes with the accounting and corpus; no general selective-risk advantage of the gate is established. Significance . Corpus labels, criterion flags and latent expert unreadability must not be conflated. A gate can win by predicting a labeling convention, and this retrospective study validates neither transfer to smartwatch dry electrodes nor improvement in patient outcomes. Measure the inconclusive rate with a stated criterion on the raw signal; do not inherit it from a corpus label.
Abstract Objective. To evaluate the transferability of interpretable single-feature threshold models across the five evaluated databases for congestive heart failure (CHF) detection from short-term heart rate variability (HRV), and to compare their performance with optimized multi-feature classifiers. Approach. Twenty HRV features from 60 s inter-beat interval windows in five databases were evaluated using leave-one-database-pair-out nested cross-validation with six outer-test sets and their corresponding outer-development sets. Within each outer-development set, directions and thresholds were fitted on inner-train sets, and features were selected according to mean inner-validation F 1 score. The selected model was refitted on the complete outer-development set. The most frequently selected feature was further evaluated across the six outer-test sets with separately fitted directions and thresholds. Eight multi-feature classifiers were optimized and evaluated using the same data splits. Performance was assessed using record-balanced window-level metrics and database-stratified recording-level bootstrap confidence intervals. Main results. The short-term detrended fluctuation analysis scaling exponent (f14) was selected most frequently. The f14 threshold model achieved the highest mean test AUC (0.881 ± 0.038), F 1 score (0.718 ± 0.079), and sensitivity among the evaluated single-feature and multi-feature models. Among the eight multi-feature classifiers, Random Forest achieved the highest mean test AUC (0.858 ± 0.067). Significance . An interpretable single-feature threshold model showed strong CHF–normal sinus rhythm discrimination in the five evaluated databases. f14 was the strongest candidate in this evaluation, although its thresholds varied across outer-development sets and its clinical specificity requires further validation.
Abstract Objective. This study aimed to improve peripheral oxygen saturation (SpO2) estimation across a wide range for hypoxemic-risk assessment in chronic obstructive pulmonary disease (COPD) using dual-wavelength photoplethysmography (PPG) and machine learning. Approach. A MAX30102-based module acquired synchronized red (660 nm) and near-infrared (880 nm) PPG signals from the right index finger. A CMS50DA oximeter on the right middle finger provided concurrent reference SpO2 readings for training and evaluation. Smartphone video captured its display for manual label extraction and independent checking. The ratio-of-ratios (R) and morphological, temporal, statistical, and derivative features were extracted. Principal component analysis-based feature selection identified a feature subset. Machine-learning and PPG-oriented deep-learning models were compared with conventional R-ratio calibration using mean absolute error (MAE), root mean squared error (RMSE), and the coefficient of determination (R2). Main results. The Extra Trees regressor achieved the lowest RMSE among evaluated models, with an MAE of 0.38 percentage points, an RMSE of 0.90 percentage points, and an R2 of 0.91. It outperformed conventional R-ratio calibration. In the hypoxemic range (70% ≤ SpO2 ≤ 90%), MAE was 0.98 percentage points, RMSE was 2.04 percentage points, and R2 was 0.66. Significance. Combining dual-wavelength PPG with multi-domain features and machine learning shows potential for wide-range SpO2 estimation and hypoxemic-risk assessment in COPD.
