Medication Monitoring from Accelerometer Data through a Series of Medication Actions Using Neural Network for Medication Adherence Evaluation

The method of medication distribution differs from country to country, and because the standard method of medication distribution in Japan is the use of PTP (Press Through Pack) sheets, it is necessary to study a recognition method that does not depend on medication bottles. Therefore, we aimed to improve medication adherence by applying a method specific to Japanese medication. As a preliminary step, in this study, we defined “medication” as the action of taking out a medication from a PTP sheet instead of a bottle and pouring into 200 ml of water, measured using a simple device attached to the wrist with a triaxial acceleration sensor, and detected the presence of medication through an artificial neural network. The motions collected were “taking medication,” “medication-like motions,” and “walking.” As a result, the overall accuracy was 94.4%, and the accuracy was 91.7% for opening the PTP sheet and placing the medication into the mouth, and 91.7% for the actions leading up to swallowing the medicine with water, all of which we prepared were recognized as medication. We were also able to show that 200 ml of water can be used to recognize the medication even when using a PTP sheet.