Smartphone-Based Noninvasive Automatic Dengue Detection Using Fingertip Video

The dengue (DNG) outbreak has created a global crisis that has raised mortality and added a burden on healthcare systems around the world. Early DNG detection is essential to prevent severe complications. However, the existing diagnostic method depends on invasive blood tests, which are costly, time–consuming, and inaccessible in low-resource areas. This paper presents a novel non-invasive smartphone-based method for DNG detection using the photoplethysmography (PPG) signal from the fingertip video. During data collection, fingertip videos were collected from 170 participants, comprising 82 normal and 88 dengue-positive cases. The raw PPG signal is extracted from the fingertip videos at a frame rate of 30 fps. This proposed method has employed advanced data acquisition and signal pre-processing techniques to enable robust feature extraction for accurate DNG detection. Fourier analysis, 1st and 2nd order derivative, and single-period analysis are used to extract 45 features from the clean PPG signals. An ensemble decision method has been employed for DNG detection, as it integrates predictions from multiple base learners to improve overall detection reliability and reduce individual model bias. Using Recursive Feature Elimination (RFE) based feature selection, the Voting Classifier has shown strong performance by combining six distinct models in differentiating DNG cases from non-cases with an accuracy of 88%. It has ensured robust precision, F1-score, sensitivity, specificity, and AUC values of 0.887 ± 0.179, 0.875 ± 0.187, 0.887 ± 0.199, 0.867 ± 0.293, and 0.94, respectively. This user-friendly smartphone-based non-invasive dengue detection method provides an alternative to conventional laboratory diagnostics for early DNG identification in resource-limited areas.