A robust motion correction technique for infrared thermography during awake craniotomy

Area of Science:

• Neurosurgery
• Medical Imaging
• Biomedical Engineering

Background:

• Intraoperative infrared thermography is emerging for image-guided neurosurgery.
• Physiological and pathological processes cause temperature changes detectable by thermography.
• Motion during data acquisition introduces artifacts, hindering accurate analysis of brain surface thermography.

Purpose of the Study:

• To develop a fast and robust technique for motion estimation and correction in brain surface thermography.
• To address artifacts caused by patient motion during intraoperative recordings.
• To enhance the preprocessing of thermal data for neurosurgical applications.

Main Methods:

• Developed a Bispline registration technique approximating motion as 2D bilinear splines.
• Incorporated a regularization function to ensure biomechanically feasible motion solutions.
• Compared Bispline registration against phase correlation, band-stop filter, Demons, Horn-Schunck, and Lucas-Kanade methods using patient data.

Main Results:

• Bispline registration demonstrated the lowest mean-squared error and highest peak-signal-to-noise ratio among tested methods.
• Performance on the structural similarity index metric was comparable to phase correlation and Demons registration.
• Band-stop filtering and Lucas-Kanade were less effective; Horn-Schunck performance degraded over time.

Conclusions:

• Bispline registration exhibited the most consistent and strong performance for nonrigid motion correction in thermography.
• The technique is computationally efficient, processing ten frames per second, suitable for real-time neurosurgical use.
• Regularization and interpolation effectively constrain deformation for fast, monomodal motion correction of thermal data.