Abstract Background and aims In around 50% of stroke survivors, dysphagia is observed as a detrimental condition. Swallowing dysfunction remains persistent in around 11-13% of stroke patients even after several months. In this study we aim at developing a wearable, non-invasive device for detecting swallowing dysfunction in stroke patients. Methods The study was initiated after Institute Ethics Committee approval and CTRI registration (CTRI/2023/11/060345). 340 patients were screened from Neurology Wards, 222 patients (65.3%) had swallowing dysfunction, while 118 patients (34.7%) had no reported swallowing issues. 10 cases and 5 controls have been recruited till date on which our developed device has been tested simultaneously with video-fluoroscopy (VFS). Both the cases and controls were stepwise given liquid (diluted barium), then semi-solid (Microbar HD) and lastly solid (breadcrumbs coated with barium). Results The collected signals were processed using the Wavelet Scattering Transform (WST), which extracted robust, time-invariant features from both modalities, capturing relevant Spectro temporal information while ensuring resilience to noise and inter-subject variability. To manage the high-dimensional feature space, dimensionality reduction was performed using Principal Component Analysis (PCA) and Multidimensional Scaling (MDS). These reduced feature sets were then used to train a Support Vector Machine (SVM) classifier to distinguish between healthy and dysphagic swallows. The SVM classifier achieved an overall accuracy of over 90%, with a sensitivity and specificity exceeding 88%, demonstrating strong agreement with VFS-based ground truth labels. Conclusions This is an ongoing trial with the main aim of determining the diagnostic accuracy of wearable swallowing dysfunction detecting device. Conflict of interest Awadh Kishor Pandit: nothing to disclose
Pandit et al. (Fri,) studied this question.