Instillation checking using long short-term memories for ophthalmology patients

In this paper, we propose a method of checking the eye lotion instillation for ophthalmology patients. Our method first estimates tilt angles of an eye dropper bottle from acceleration values measured by a triaxial sensor attached to the bottle. It next prepares data each of which is equal to a sequence of standardized tilt values, as data to be presented to a discrimination model. It employs either a long short-term memory (LSTM for short) or a bidirectional long short-term memory (B_LSTM for short) to construct the model. Once we present the data to be checked to our model, it produces a certainty degree indicating whether a patient corresponding to the presented data applies eye lotion at the time zone in which a sequence of the tilt values used to prepare the presented data was measured. The final judgement for the instillation depends on thresholding of the certainty degree. It is established, from experimental results using practical data, that adopting B_LSTM-based models is useful in improving metric values compared to adopting LSTM-based models, and that our models can cope well with the situation specified by the comparatively long time interval of measuring tilt values while some metric values slightly degrade.