Diabetic retinopathy (DR) is a medical condition caused by diabetes. The development of retinopathy significantly depends on how long a person has had diabetes. Initially, there may be no symptoms or just a slight vision problem due to impairment of the retinal blood vessels. Later, it may lead to blindness. Recognizing the early clinical signs of DR is very important for intervening in and effectively treating DR. Thus, regular eye check-ups are necessary to direct the person to a doctor for a comprehensive ocular examination and treatment as soon as possible to avoid permanent vision loss. Nevertheless, due to limited resources, it is not feasible for screening. As a result, emerging technologies, such as artificial intelligence, for the automatic detection and classification of DR are alternative screening methodologies and thereby make the system cost-effective. People have been working on artificial-intelligence-based technologies to detect and analyze DR in recent years. This study aimed to investigate different machine learning styles that are chosen for diagnosing retinopathy. Thus, a bibliometric analysis was systematically done to discover different machine learning styles for detecting diabetic retinopathy. The data were exported from popular databases, namely, Web of Science (WoS) and Scopus. These data were analyzed using Biblioshiny and VOSviewer in terms of publications, top countries, sources, subject area, top authors, trend topics, co-occurrences, thematic evolution, factorial map, citation analysis, etc., which form the base for researchers to identify the research gaps in diabetic retinopathy detection and classification.

The blood pressure (BP) estimation plays a crucial role in assessing cardiovascular health and preventing related complications. One of the early warning indicators for heart disorders is elevated blood pressure. Thus, monitoring of blood pressure continuously is needed. The study aims to develop and validate a reliable deep learning-based approach for blood pressure estimation using photoplethysmography from the publicly available database MIMIC-II. The continuous wavelet transform (CWT) was used to transform the photoplethysmogram (PPG) signals into scalograms, which were then input into six different deep learning models: VGG16, ResNet50, InceptionV3, NASNetLarge, InceptionResNetV2 and ConvNeXtTiny. The obtained deep features from each one of these models were employed to estimate BP values using random forest. The proposed approach uses a unique transfer learning framework that integrates deep feature extraction from scalograms with random forest regression, providing a new pathway for blood pressure estimation. The models were assessed using mean absolute error (MAE) and standard deviation (SD) in estimating the systolic and diastolic blood pressure values. Out of six models, ConvNeXtTiny and VGG16 showed good performance. ConvNeXtTiny achieved mean absolute error of 2.95 mmHg and standard deviation of 4.11 mmHg for systolic blood pressure and mean absolute error of 1.66 mmHg and standard deviation of 2.60 mmHg for diastolic blood pressure. The achieved result complies with the clinical standards set by Advancement of Medical Instrumentation Standard (AAMI) and the British Hypertension Society standard (BHS). This can enhance cardiovascular health monitoring with continuous, non-invasive and reliable blood pressure measurement, assisting in early detection of the disease. The suggested method shows that reliable blood pressure estimation from photoplethysmography signals is possible with the use of deep learning and transfer learning. Above all, ConvNeXtTiny offers a dependable method for continuous blood pressure monitoring that satisfies clinical requirements and may help in the early identification of cardiovascular problems.

The increasing prevalence of cardiovascular diseases demands affordable, accessible, and continuous cardiac monitoring solutions. Although an Electrocardiogram (ECG) is considered the clinical gold standard, long-term real-time monitoring of healthy individuals is limited. On the other hand, photoplethysmography is a simple and cost-effective technique, but it can be susceptible to noise and signal distortion. This study attempts to bridge the gap by proposing a novel framework for Photoplethysmogram (PPG) to ECG transformation, using spectral transformation methods such as Short-Time Fourier Transform (STFT), Discrete Cosine Transform (DCT), Wavelet Transform (WT), and Fast Fourier Transform (FFT). Deep learning models, namely Convolutional Neural Network (CNN), Convolutional Neural Network-Gated Recurrent Unit (CNN-GRU), and Convolutional Neural Network-Long Short-Term Memory (CNN-LSTM), were trained on the UCI Machine Learning Repository version of the MIMIC II dataset and PulseDB Vital datasets to reconstruct ECG from PPG, achieving a correlation coefficient of 0.9731. Validated on MIMIC III, robustness was confirmed with a correlation of 0.9639. The proposed framework leverages frequency-domain representations and lightweight, efficient deep learning models, making it well-suited for integration into real-time and resource-limited environments, such as wearable health monitoring systems, and serves as a foundation for future non-invasive blood pressure estimation using only PPG input.