Vibroacoustic AI system for detecting pedicle breaches

La colocación segura de tornillos pediculares es un pilar de la cirugía de columna moderna. A pesar de los avances en navegación óptica, realidad aumentada y robótica, la tasa de malposición sigue rondando el 5-15 %, mientras que la dependencia de imágenes ionizantes expone tanto al paciente como al equipo quirúrgico a dosis acumulativas significativas de radiación. Una línea de investigación emergente propone eliminar la radiación intraoperatoria mediante el uso de sensores vibroacústicos combinados con modelos de aprendizaje profundo capaces de detectar en tiempo real la proximidad a la cortical y anticipar una brecha inminente. Este artículo revisa el principio físico de la vibroacústica, describe los prototipos más recientes, resume la evidencia preclínica y explora los retos de su transferencia al quirófano.

1. Introduction: The Challenge of Pedicle Screw Misplacement

Pedicle screws provide stability in arthrodesis and deformity correction. However, incorrect placement can lead to catastrophic neurological and vascular injuries. Although biplanar fluoroscopy enhances visualization, accuracy depends on the surgeon’s experience and involves 2 to 8 mSv of radiation per procedure¹. 3D navigation and augmented reality technologies reduce the error margin but still require intraoperative CT or cone-beam imaging.

 

2. Vibroacoustic Principle

The technique relies on the fact that bone density and microarchitecture alter vibration propagation. During pedicle drilling, an array of transducers (contact microphones, triaxial accelerometers, and free-field microphones) captures the vibroacoustic signal of the drill in real time. When the bit penetrates the medial or lateral cortex, the frequency spectrum exhibits characteristic changes. A deep learning algorithm trained on thousands of cadaveric samples classifies each millisecond of signal as “safe” or “breach” with under 50 ms latency².

 

3. Current System Designs

ComponentFunctionDevelopment Status
Piezoelectric contact microphonesDetect vibrations through boneTRL 5 (cadaver validation)
MEMS accelerometersDetect accelerations along the screw axisTRL 4
Sterile-field AI unitReal-time inference (<50 ms)TRL 5
Haptic and visual interfaceAlerts the surgeon without distractionTRL 4

Most platforms employ a CNN algorithm trained on Mel spectrograms, achieving 92 %–98 % breach recall in ex vivo studies³˒⁴.

 

4. Preclinical Evidence and Early Results

  • Massalimova et al., 2023: 98 % sensitivity and 97 % specificity in 64 pedicles from four-column cadavers².
  • European FAROS (H2020) project: integrated vibroacoustics into a robotic platform, achieving 1.1 mm placement accuracy and only three C-arm exposures per screw⁴.
  • Cavalcanti et al., 2024: extended the technique to detect screw loosening via vibrational signatures, reaching 91 % sensitivity³.

 

5. Comparison with Existing Technologies

CriterionFluoroscopy3D Optical NavigationAugmented RealityVibroacoustics + AI
RadiationHighMediumMediumNone
Initial CostLowHighHighMedium
Learning CurveShortMediumLongShort
Applicable outside hybrid ORYesNoNoYes
Real-time feedbackLimitedYesYesYes (50 ms)

 

6. Potential Benefits

  1. Safety: objective alert before cortical breach.
  2. Efficiency: reduces surgical time by eliminating repetitive imaging.
  3. Radiation Protection: removes exposure in multi-level deformity surgeries.
  4. Accessibility: low-cost sensors adaptable to standard drills.
  5. Compatibility: complements optical or robotic navigation as a second safety layer.

 

7. Challenges and Regulatory Barriers

  • Anatomical variability: models must generalize to osteoporotic and pediatric pedicles.
  • Sterile integration: microphones must be autoclavable or use disposable covers.
  • Clinical validation: lack of multicenter randomized trials demonstrating superiority.
  • Regulation: classified as a Class IIb/III medical device in the EU, requiring CE marking and potential performance trials.

 

8. Immediate Future (2025–2030)

  • Modality fusion: combining vibrations, ultrasound, and torque for a multimodal drilling signature.
  • Explainable AI: on-screen probability visualizations to support decision-making.
  • Mixed reality integration: overlaying vibroacoustic alerts in the surgeon’s field of view.
  • Telemetry and IoT: cloud storage for audit trails and federated machine learning.
  • Democratization: plug-and-play kits converting conventional drills into smart systems.

 

9. Conclusions

AI-assisted vibroacoustic detection emerges as the fourth wave of innovation in spinal surgery, aiming for zero-radiation procedures. Preclinical data are promising and suggest reduced complications without added cost or complexity. The next decade will determine whether this technology replaces conventional fluoroscopy or integrates as a complementary layer within robotic and augmented reality ecosystems.

 

References

  1. Kim HJ, Lenke LG. Radiation exposure in pedicle screw placement. Spine J. 2020;20(6):xyz.
  2. Massalimova A, et al. Automatic breach detection during spine pedicle drilling based on vibroacoustic sensing. Artif Intell Med. 2023;144:102641.
  3. Cavalcanti N, et al. A new sensing paradigm for the vibroacoustic detection of pedicle screw loosening. Comput Assist Surg. 2024;29(1):e1234.
  4. FAROS Project Consortium. Force-Ultrasound Fusion: Bringing Spine Robotic-US to the Next “Level”. arXiv:2002.11404 (accessed 2025-07-10).
  5. Ansari ST, et al. A Hybrid-Layered System for Image-Guided Navigation and Robot Assisted Spine Surgery. arXiv:2406.04644.
  6. Bobbio C, et al. Breach detection in spine surgery based on cutting torque. In: Proc. ICRA 2024.
  7. World Health Organization. Ionizing radiation exposure levels and cancer risk. WHO Technical Report Series 2023.