Risk assessment of diabetic retinopathy with machine and deep learning models with PPG signals and PWV.

Retinopathy is one of the most common micro vascular impairments in diabetic subjects. Elevated blood glucose leads to capillary occlusion, provoking the uncontrolled increase in local growth of new vessels in the retina. When left untreated, it can lead to blindness. Traditional approaches for retinopathy detection require expensive devices and high specialized personnel. Being a microvascular complication, the retinopathy could be detected using the photoplethysmography (PPG) technology. In this paper we investigate the predictive value of the pulse wave velocity and PPG signal analysis with machine and deep learning approaches to detect retinopathy in diabetic subjects. PPG signals and pulse wave velocity (PWV) showed promising results in assessing the diabetic retinopathy. The best performances were scored by a LightGBM based model trained over a subset of the available dataset obtaining 80% of specificity and sensitivity.Clinical relevance- PPG based retinopathy detection could make the retinopathy detection more accessible since it does not need neither expensive devices for signal acquisition nor highly specialized personnel to be interpreted.

Clustered photoplethysmogram pulse wave shapes and their associations with clinical data.

Photopletysmography (PPG) is a non-invasive and well known technology that enables the recording of the digital volume pulse (DVP). Although PPG is largely employed in research, several aspects remain unknown. One of these is represented by the lack of information about how many waveform classes best express the variability in shape. In the literature, it is common to classify DVPs into four classes based on the dicrotic notch position. However, when working with real data, labelling waveforms with one of these four classes is no longer straightforward and may be challenging. The correct identification of the DVP shape could enhance the precision and the reliability of the extracted bio markers. In this work we proposed unsupervised machine learning and deep learning approaches to overcome the data labelling limitations. Concretely we performed a K-medoids based clustering that takes as input 1) DVP handcrafted features, 2) similarity matrix computed with the Derivative Dynamic Time Warping and 3) DVP features extracted from a CNN AutoEncoder. All the cited methods have been tested first by imposing four medoids representative of the Dawber classes, and after by automatically searching four clusters. We then searched the optimal number of clusters for each method using silhouette score, the prediction strength and inertia. To validate the proposed approaches we analyse the dissimilarities in the clinical data related to obtained clusters.

Assessment of the systolic rise time by photoplethysmography in peripheral arterial diseases: a comparative study with ultrasound Doppler.

Aims: Peripheral arterial disease (PAD) is a major public health burden requiring more intensive population screening. Ankle brachial index (ABI) using arm and ankle cuffs is considered as the reference method for the detection of PAD. Although it requires a rigorous methodology by trained operators, it remains time-consuming and more technically difficult in patients with diabetes due to mediacalcosis. Techniques based on the study of hemodynamic, such as the systolic rise time (SRT), appear promising but need to be validated. We retrospectively compared the reliability and accuracy of SRT using a photoplethysmography (PPG) technique to the SRT measured by ultrasound doppler (UD) in PAD patients diagnosed with the ABI (137 patients, 200 lower limbs). Methods and results: There was a significant correlation between SRT measured with UD (SRTud) compared with that with PPG (SRTppg, r = 0.25; P = 0.001). Best correlation was found in patients without diabetes (r = 0.40; P = 0.001). Bland and Altman analysis showed a good agreement between the SRTud and SRTppg. In contrast, there was no significant correlation between UD and PPG in diabetes patients. Furthermore, patients with diabetes exhibited a significant increase of SRTppg (P = 0.02) compared with patients without diabates but not with the SRTud (P = 0.18). The SRTppg was significantly linked to the arterial velocity waveforms, the type of arterial lesion but not vascular surgery revascularization technique. Conclusion: This monocentric pilot study shows that SRT measured with the PPG signal reliably correlates with SRT recorded with UD. The PPG is an easy to use technique in the hand of non-expert with a potential interest for general screening of PAD, especially in diabetes patients, due to its ease to use.