Dentists still consider brushing the most reliable way to prevent cavities and gum disease. This publication has previously introduced a self-test kit that measures how far gum inflammation has already progressed, a technology that gauges whether disease is already underway. A study recently published by a research team at National Yang Ming Chiao Tung University in Taiwan measures an earlier stage: how evenly the brushing itself, which prevents cavities and gum disease, is actually carried out. Using two small sensors attached to the wrist and the toothbrush, the method reads which of up to 18 regions of the mouth is being brushed at any given moment. Even though it is a preliminary study based on a single participant's data, its accuracy exceeded 96%.
Two Inertial Sensors Track the Toothbrush's Position
This technology reads signals from two inertial sensors, devices that measure movement through acceleration and rotational speed. One sits inside a holder that clips onto the toothbrush handle, and the other is worn on the wrist, the same principle a smartphone uses to automatically rotate its screen. The research team also calculated Euler angles, values that indicate the angle at which the toothbrush is currently tilted. This is because regions that are hard to distinguish from wrist movement alone, such as the inner surface of the upper front teeth versus the inner surface of the lower front teeth, can be told apart once you know whether the brush is pointed up or down. When the team fed the combined signals into six machine learning models for comparison, a random forest model, which builds multiple decision criteria and reaches a conclusion by majority vote, proved the most accurate.
A Wrist Sensor Alone Was Not Enough
Most existing smart toothbrushes are sold with a sensor built directly into the handle. But since a toothbrush is a consumable that must be replaced every two or three months, this effectively means throwing away an expensive sensor along with it each time. This study instead fitted a commercially available wearable sensor into an inexpensive 3D-printed holder, so it can be attached to any ordinary toothbrush. Using only the sensor on the toothbrush kept accuracy at around 95%, while using only the wrist-worn sensor dropped it to 82%. That is because the wrist captures the direction the arm is moving but cannot pick up the fine rotations of the brush made with the fingers in the mouth's tight space. Accuracy exceeded 96% only when both sensors were used together.
Filtering Out Transition Segments Reveals the Actual Brushing Time
Prior studies that classified brushing regions typically counted the brief moments spent moving from one region to another as brushing time as well. That inflates the measured brushing time with periods when no actual brushing occurred, making coverage look more even than it really is. This study first separated the segments where the brush is moving from the segments where it is actually brushing a surface, and only then identified which of the 18 regions was being brushed. Errors that had confused left-right symmetric regions mostly disappeared once Euler angles were added. However, two closely adjacent regions, the chewing surface and the inner surface on the lower left side, were still confused with each other, with accuracy reaching only 83% and 91% respectively.
Lab Accuracy and Real-World Use Are Different Matters
This study is a preliminary one, built from data collected over 30 brushing sessions across two weeks from a single participant. Since hand size and habits vary from person to person, it remains unconfirmed whether the same accuracy would hold across a larger population. In fact, in a trial where Colgate-Palmolive distributed smart connected toothbrushes to 409 elementary school students in the UK, 29% of participants never connected the device and app at all. Even among the children who did connect successfully, fewer than half brushed twice a day on any given day. This shows that even with technology that accurately identifies brushing regions, getting children to use the device consistently is a separate challenge altogether.
Two inexpensive sensors attached to the toothbrush handle and the wrist, combined with a machine learning model that adds directional information, have reached the point of distinguishing up to 18 brushing regions. This is a level of precision that toothbrushes with expensive built-in sensors, or a wrist sensor alone, could not previously achieve. The next challenge is to reproduce this precision across a larger population and to get people to actually keep using the device.