BACKGROUND
Dysphagia is a rapidly progressive disease mainly caused by stroke, dementia, traumatic brain injuries, and Parkinson’s disease. It is defined as a condition in which food material does not enter the esophagus, instead getting caught or aspirated into the trachea. In an aging society, the number of patients with dysphagia who require proper treatment and management is expected to increase; when food material gets aspirated into the trachea, it can cause aspiration pneumonia and eventually lead to death in the elderly or patients with neurological diseases.
OBJECTIVE
In general, normal swallowing, which can be assessed using videos, is classified into three phases: oral, pharyngeal, and esophageal; and dysphagia is classified mainly into two categories: penetration and aspiration. This study proposes a video analysis-based artificial intelligence (AI) web application that diagnoses dysphagia by labeling video files and applying an AI model.
METHODS
The videofluoroscopic swallowing study (VFSS) is the most commonly used test to diagnose dysphagia in elderly patients. Regarding VFSS exam files, they are multiframe data that contain 200–700 images. Therefore, they can be very large, ranging from 300 MB to 1.5 GB in size. In the study, to label the data, the server separated them into frames during the upload and stored them as a video for analysis. Then, the separated data were loaded into a labeling tool to perform labeling. The labeled file was downloaded, and an AI model was developed by training with You Only Look Once (YOLOv7).
RESULTS
When a VFSS video file was uploaded to an application equipped with the developed AI model, it was automatically classified and labeled as either oral, pharyngeal, or esophageal. The dysphagia of a person was categorized as either penetration or aspiration; the final analyzed result was displayed to the viewer. The following labeling datasets were created for AI learning: oral (n = 2355), pharyngeal (n = 2338), esophageal (n = 1480), penetration (n = 1856), and aspiration (n = 1320); the learning results of the YOLO model, which analyzed dysphagia using the dataset, were predicted with accuracies of 0.90, 0.82, 0.79, 0.92, and 0.96, respectively.
CONCLUSIONS
Although conducting the VFSS is simple, carrying out the preparation for the test is time-consuming and labor-intensive. Patients who need to take it frequently, such as those with stroke, dementia, and Parkinson's disease, need to travel to a tertiary general hospital, which is difficult considering their restricted mobility and the long waiting time. We expect that the proposed study can support clinical practices in simplifying this inconvenient process and helping clinicians with reading tasks.
CLINICALTRIAL
N/A