Deep Learning-Based Fine-Tuning Approach of Coarse Registration for Ear-Nose-Throat (ENT) Surgical Navigation...

Area of Science:

• Medical Imaging
• Computer-Aided Surgery
• Machine Learning

Background:

• Accurate medical image registration is vital for surgical navigation in minimally invasive procedures.
• Current methods may have limitations in precision, especially in complex anatomical regions.

Purpose of the Study:

• To introduce a novel deep learning-based refinement step to improve surface registration accuracy.
• To integrate this method into existing surgical navigation workflows without disruption.

Main Methods:

• A deep learning model was trained on simulated anatomical landmarks with localization errors.
• The model utilizes global feature learning, iterative prediction, and independent rotation/translation processing.
• Validation involved silicon-masked head phantoms and CT imaging, comparing against conventional and prior deep learning approaches.

Main Results:

• The proposed method significantly reduced target registration error (TRE) compared to conventional and previous deep learning methods (1.58 ± 0.52 mm vs. 2.37 ± 1.14 mm and 2.29 ± 0.95 mm, respectively).
• Improved accuracy was observed across various facial regions and depths.
• Consistent performance was noted, particularly enhancing registration for deeper anatomical areas.

Conclusions:

• The novel deep learning refinement step enhances surface registration accuracy for surgical navigation.
• This advancement offers potential for increased precision and safety in minimally invasive surgeries.
• The method integrates seamlessly into established workflows, offering practical clinical benefits.