Research On Prescription Recommendation Based On Tongue Image Retrieval

Traditional Chinese Medicine(TCM)is an important part of China’s healthcare system.The country has issued a series of policies to develop TCM.In all human organs,the tongue contains rich diagnostic information,which is an important reference for doctors when prescribing.As the easiest organ for data collection,the tongue is the most suitable one to open the intelligent research of TCM.The study of prescription recommendation method based on tongue image can provide reference for doctors and facilitate the follow-up research on the intelligence of TCM.This paper recommends prescriptions based on tongue images to provide reference for doctors,and this paper:(1)proposes three prescription recommendation methods and develops several metrics to verifies their performances.(2)studies the effect of dose information on prescription recommendation and evaluates the performance of prescription recommendation by experts.(3)proposes the extraction method of multi-grained tongue feature and proposes the grouping constraints to guarantee the consistency and smoothness of grouping at different granularity to enhance prescription recommendation.The results of experiments on clinical data sets,including prescription generation,prescription retrieval and tongue image retrieval,show that the method of tongue image retrieval is most effective.Combined with the introduction of dose,sample weighting and tongue feature extraction method proposed in this paper,the prescription recommendation improved 7.26% on the NDCG-iou.The experimental results show that(1)it is more difficult to rely only on the tongue image to generate prescriptions,and it is more appropriate to use the tongue image as a filter to filter prescriptions in the database;(2)prescription feature extraction needs the support of big data,which is not feasible at present,and(3)the introduction of dose information,sample weighting,and improving the extraction method of tongue feature are all effective strategies to improve the performance of prescription recommendation.