Author(s)
Tsuyoshi Kojima1
2
Shintaro Fujimura1
Kengo Oe1
Yusuke Okanoue2
Shuya Otsuki2
Affiliation(s)
1 Kyoto University, 2 Tenri Hospital;
Abstract:
Objective: A fiberoptic endoscopic evaluation of swallowing (FEES) is an essential examination of swallowing function. Although the evaluation method using scores is useful for daily clinical use for FEES, it is difficult to capture detailed changes.
Method: We have developed a method to analyze FEES images using artificial intelligence (AI) to evaluate the information available in FEES as objectively and in as much detail as possible. First, the anatomical location, residual swallows, and salivary retention were learned by color-coding and labeling the still images from the FEES videos. Then, the learned model obtained from the learning of still images was applied to each FEES movie. Finally, the learned models from the still images were used to infer each FEES video frame (30 fps).
Results: The AI could recognize anatomical structures, colored water, saliva, etc., in the video by using the learned models from the still images to infer each frame of the FEES video. The changes in anatomical structures, such as the epiglottis and pear-shaped depression over time, can be expressed in terms of the timing of the swallowing reflex, whiteout, etc. In addition, the changes in the volume of swallowed material can be used to quantitatively evaluate the amount of swallowed material remaining in the mouth or repeated swallowing.
Conclusions: The detailed evaluation of the swallowing disorder could lead to the selection of optimal food forms and the determination of rehabilitation's effectiveness by patterning the swallowing disorder, thus improving the quality of the rehabilitation process.