Machine learning model for fast prediction and uncertainty quantification of needle deflection during prostate biopsy

Area of Science:

• Medical Engineering
• Computational Mechanics
• Surgical Planning

Background:

• Transperineal prostate biopsies are gaining interest due to reduced infection risk compared to transrectal approaches.
• Accurate needle placement is critical but challenging in transperineal biopsies due to long insertion distances and potential tissue property variations.
• Predictive models for needle deflection can improve targeting accuracy and reduce procedure attempts.

Purpose of the Study:

• Develop a computationally efficient model for predicting biopsy needle deflection suitable for intraoperative planning.
• Validate the predictive model against experimental data.
• Integrate the model into a Monte Carlo uncertainty quantification framework to assess the impact of tissue property variations.

Main Methods:

• A mechanics-based model of needle deflection was used to train a Fourier Feature Neural Network (FFNN).
• Both the mechanics-based model and the FFNN were validated using experimental data from tissue phantoms.
• The FFNN model was employed within a Monte Carlo simulation for uncertainty quantification of needle deflection.

Main Results:

• Both mechanics-based and FFNN models demonstrated good agreement with experimental results.
• The FFNN model achieved high accuracy, with a tip deflection error of approximately 0.3 mm compared to the mechanics-based model.
• A low-computational-cost Monte Carlo uncertainty quantification model (approx. 20 CPU seconds) was demonstrated, showing the influence of tissue depth, stiffness, and uncertainty on needle deflection.

Conclusions:

• A computationally efficient Monte Carlo uncertainty quantification model for needle deflection was successfully developed.
• This approach shows significant potential for enhancing intraoperative planning in prostate biopsies.
• The method is also applicable to other transperineal procedures involving flexible needles, such as cryoablation and brachytherapy.