Percutaneous nephrostomy guidance by a convolutional-neural-network-based optical coherence tomography endoscope.

Area of Science:

• Urology
• Medical Imaging
• Surgical Navigation

Background:

• Percutaneous nephrostomy is a common urological intervention.
• Challenges include precise needle placement and avoiding hemorrhage from blood vessel rupture.

Purpose of the Study:

• To develop an endoscopic optical coherence tomography (OCT) probe for enhanced needle navigation during percutaneous nephrostomy.
• To improve the safety and accuracy of kidney access surgeries.

Main Methods:

• Development of an endoscopic OCT probe for real-time imaging.
• Experiments conducted on thirty-one human kidneys for tissue and blood vessel analysis.
• Utilization of deep learning models (Inception and nnU-net) for automated recognition.

Main Results:

• Effective distinction of renal tissues (cortex, medulla, calyx, sinus fat, pelvis) using structural OCT imaging.
• Successful detection of renal blood flow via OCT Doppler function.
• High accuracy achieved: 99.6% for tissue classification (Inception) and 0.8917 IoU for blood vessel detection (nnU-net).

Conclusions:

• Endoscopic OCT probe enables precise identification of renal structures and blood vessels.
• Deep learning integration automates recognition, enhancing procedural safety.
• This technology offers significant potential to improve percutaneous nephrostomy outcomes.