Projects

Artificial Intelligence-based Video fluoroscopy Analysis for a Deglutology Decision Support System

The video fluoroscopic swallowing study (VFSS) is the gold standard for assessing dysphagia. It is an x-ray imaging modality that records the swallowing procedure of patients with different amounts of food consistency. Dysphagia is a swallowing disorder combined with physiologic impairments that occur during eating and drinking. The human swallowing process contains three phases: oral, pharyngeal, and esophageal. The pharyngeal swallowing assessment in VFSS plays a crucial role in assessing the presence and severity of dysphagia, as well as the risk of aspiration and penetration. A pharyngeal swallowing assessment can also determine the optimal bolus volume, viscosity, and texture for safe and efficient swallowing and the need for compensatory strategies or maneuvers. The Modified Barium Swallow Impairment Profile (MBSImP) is a research-based standardized clinical rating tool used for the VFSS interpretation and diagnosis of dysphagia. It is a qualitative assessment, requiring clinicians to use an element of subjectivity. Artificial Intelligence (AI) is used as a promising tool in the medical field for reducing human errors. Applying AI in VFSS assessments can bring several benefits: AI can analyze massive amounts of data more efficiently and precisely. AI can help standardize the VFSS procedure, reducing the variability observed in clinical practice. AI can help overcome human biases during evaluation. The AI with VFSS analysis study on the Indian population has the potential improvement.

The initial pilot study was conducted to detect the penetration-aspiration risk from VFSS videos. In dysphagia patients, the invasion of food material to the airway, called penetration or aspiration, can lead to asphyxia or pneumonia risk. The penetration or aspiration risk assessment from the VFSS using the penetration-aspiration scale is known to be subjective,

with a high degree of inter-rater and inter-patient variability. We developed a clinical tool called SPAD (Swallowing phase classification coupled Penetration-Aspiration Detection) for classifying the different phases of swallowing from VFSS videos and subsequent identification of penetration-aspiration by assessment of pharyngeal bolus residue, which has a positive correlation with penetration-aspiration risk. SPAD tool demonstrates an Fl Score of 0.75 and 0.79 on phase classification and penetration-aspiration risk detection, respectively. Furthermore, the risk detection is interpretably overlaid on VFSS videos using Grad-Cam, helping deglutologists better assess dysphagia.

Proposal : Inspire proposal – Artificial Intelligence-based Video fluoroscopy Analysis for a Deglutology Decision Support System. Submitted.

Future Works

1. The development of an automated end to end diagnostic pipeline for penetration-aspiration assessment from VFSS videos.
2. The development of an automated AI pipeline for assessing the pharyngeal dysphagia. This pipeline will evaluate the physiological activity of the different pharyngeal phase components.