Deep-learning-aided forward optical coherence tomography endoscope for percutaneous nephrostomy guidance.

Area of Science:

• Medical Imaging
• Surgical Technology
• Artificial Intelligence in Medicine

Background:

• Percutaneous renal access is crucial for kidney procedures, demanding improved methods for precision and reduced patient morbidity.
• Current techniques for renal access lack real-time imaging guidance for differentiating critical tissue structures.

Purpose of the Study:

• To develop and validate a forward-view optical coherence tomography (OCT) endoscopic system for percutaneous nephrostomy (PCN) guidance.
• To create a deep-learning-based platform for automatic classification of renal tissue types from OCT images, aiding surgical navigation.

Main Methods:

• An OCT endoscopic system was designed and tested on porcine kidneys to assess its imaging capabilities.
• Convolutional neural networks (CNNs), including ResNet34, MobileNetv2, and ResNet50 architectures, were trained on labeled OCT images.
• Nested cross-validation was employed for performance benchmarking and uncertainty quantification across multiple kidney samples.

Main Results:

• The OCT system successfully distinguished between renal cortex, medulla, and calyx tissues in porcine kidneys.
• ResNet50-based CNN models achieved an average classification accuracy of 82.6%±3.0% for renal tissue types.
• High precisions were noted for calyx (91%±5%) and medulla (85%±6%), with robust recall rates for medulla (91%±4%) and calyx (89%±3%).

Conclusions:

• The developed OCT endoscopic system is technically feasible for real-time guidance in PCN surgery.
• The deep learning platform demonstrates robust performance in automatically identifying renal tissue structures, potentially improving surgical accuracy and safety.
• This integrated imaging and AI approach offers a promising solution to enhance percutaneous renal access procedures.