Sensors (Basel, Switzerland) · 2026
Feasibility of Drill-Tip Position Estimation During Cortical Bone Drilling Using Force and Torque Signals.
СтатьяНазвание и аннотация: Перевод в очереди
Обработка статьи · этапы, ошибки и управление
Загрузка состояния…
Ошибки и ограничения
История · было → стало
Состояние обновляется каждые 5 секунд, пока блок открыт. Процент показывается только там, где известен объём работы. Пауза останавливает продолжение этапа; текущий запрос к источнику может завершиться.
Imai Hirotatsu, Wang Han, Kishimoto Koki, Kita Kosuke, Suzuki Yuki, Hosozawa Koki, Kanie Yuya, Furuya Masayuki, Enomoto Toshiyuki, Okada Seiji, Fujimori Takahito
Аннотация
PURPOSE: Excessive drill advancement after cortical breakthrough is a potential safety concern in orthopaedic procedures. We developed a data-driven approach to estimate the drill-tip position relative to the far cortex prior to breakthrough using time-series thrust force and spindle torque signals.
METHODS: Drilling experiments were performed on 268 porcine cortical bone specimens at a constant feed rate of 0.5 mm/s. A long short-term memory network was trained to estimate the drill-tip position from filtered force and torque signals. The reference position was derived from breakthrough timing confirmed by high-speed imaging and the programmed feed rate. Performance was evaluated using mean absolute error within the -2 to +2 mm peri-breakthrough interval. Two post hoc analyses examined whether model performance exceeded an elapsed-time baseline and whether pre-breakthrough force patterns were more consistent when expressed relative to breakthrough position than to drilling onset time.
RESULTS: The combined-input LSTM achieved an MAE of 0.20 mm, compared with 0.23 mm for force alone and 0.24 mm for torque alone. Among the representative architectures evaluated, LSTM showed the lowest regression error. A signal-blind time-only baseline yielded an MAE of 0.54 mm. The association between cortical thickness and force-decline onset was weaker when expressed in spatial coordinates relative to breakthrough than when expressed as time from drilling onset (R2 = 23% vs. 74%). These findings suggest that force and torque signals contained information associated with proximity to breakthrough beyond that provided by average drilling duration alone.
CONCLUSION: Converting sensor-derived resistance patterns into spatially anchored positional information may support proactive strategies such as controlled deceleration before penetration. The proposed approach represents a step toward exemplifying the emerging concept of surgeon-assisting Physical AI.
METHODS: Drilling experiments were performed on 268 porcine cortical bone specimens at a constant feed rate of 0.5 mm/s. A long short-term memory network was trained to estimate the drill-tip position from filtered force and torque signals. The reference position was derived from breakthrough timing confirmed by high-speed imaging and the programmed feed rate. Performance was evaluated using mean absolute error within the -2 to +2 mm peri-breakthrough interval. Two post hoc analyses examined whether model performance exceeded an elapsed-time baseline and whether pre-breakthrough force patterns were more consistent when expressed relative to breakthrough position than to drilling onset time.
RESULTS: The combined-input LSTM achieved an MAE of 0.20 mm, compared with 0.23 mm for force alone and 0.24 mm for torque alone. Among the representative architectures evaluated, LSTM showed the lowest regression error. A signal-blind time-only baseline yielded an MAE of 0.54 mm. The association between cortical thickness and force-decline onset was weaker when expressed in spatial coordinates relative to breakthrough than when expressed as time from drilling onset (R2 = 23% vs. 74%). These findings suggest that force and torque signals contained information associated with proximity to breakthrough beyond that provided by average drilling duration alone.
CONCLUSION: Converting sensor-derived resistance patterns into spatially anchored positional information may support proactive strategies such as controlled deceleration before penetration. The proposed approach represents a step toward exemplifying the emerging concept of surgeon-assisting Physical AI.