Assisting Decision-making Optimization Method And System For BPPV Treatment Based On Reinforcement Learning

Benign paroxysmal positional vertigo(BPPV)is a clinically common inner ear disorder that causes vertigo,and accurate and effective treatment of it is of great significance.At present,repositioning procedure is used to treat BPPV in clinical practice.The purpose is to move the otoliths along the semicircular canals(SCC)and enter the utricle.But due to the small size of otoliths,the trajectory cannot be directly observed,thus the accuracy of the operation is not high enough and its adaptability to individual variance is limited.Existing studies have mostly focused on modeling the influence of otoliths on the cupula and conducting in vitro animal experiments.This paper is to study the movement law of otoliths in the SCC of patients with BPPV,which is an important basis for analyzing the pathogenesis of BPPV,proving the principle of otolith repositioning procedure,and improving the therapeutic effect of BPPV.The main work of this paper is as follows.1.Improvement of the dynamic model of otoliths.According to the force characteristics of otoliths in different area of the SCC of human,the dynamic model of a single otolith in the SCC is proposed,which is helpful for understanding the movement law of otoliths that cannot be directly observed,and analyzing the principle of otolith repositioning procedure.Through experiments,it is clear that the key parameter that affects the treatent effect is the pause interval T between each rotation operation,which can be used as the theoretical basis for ensuring sufficient pause time during clinical treatment.2.Proposal of a BPPV treatment-assisted decision-making method based on reinforcement learning.The dynamic model of otoliths is used as the environment of reinforcement learning,and the model-free Sarsa and Q-learning algorithm are used to generate the motion instructions of the SCC rotation.Compared with the conventional manual operation,the reinforcement learning method can adapt to the differences of individual SCC structures,and by setting reasonable rewards,the repositioning procedure can be completed faster and smoother.Finally,the simulation experiment proved that the method based on reinforcement learning is feasible and efficient,in the otolith repositioning scene.When the interval between each decision is greater than 8 seconds,the strategy can converge.For different SCC structures and initial positions of otoliths,the purpose of otolith reposition can be achieved.3.Design of a BPPV treatment aid decision-making system based on reinforcement learning.This paper combines the traditional repositioning chair with reinforcement learning.The dynamics model of otolith gives the reference trajectory of the otolith movement in the real human body,the action output of reinforcement learning algorithm is recorded during its interaction with the model,as rotating instructions to the reposition chair,and the nystagmus infers the direction of movement of the otoliths in the SCC in real time.This system takes advantage of the reinforcement learning algorithm that can learn sequence decision making in the process of interacting with the environment,and the mechanical device can be precisely positioned.The personalized treatment plan is adapted to the structural characteristics of the patient's inner ear.The complete repositioning operation performed by the mechanical device is easy for the patient to tolerate.The ultimate goal is to quickly repositioning the otoliths and successfully treat BPPV.