Deep Learning for Autonomous Surgical Guidance Using 3-Dimensional Images From Forward-Viewing Endoscopic Optical...

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Surgical Technology

Background:

• Percutaneous nephrostomy requires precise guidance for optimal outcomes.
• Volumetric optical coherence tomography (OCT) provides detailed anatomical information.
• Current image analysis methods may limit real-time surgical decision-making.

Purpose of the Study:

• To develop and evaluate a 3D-CNN for analyzing volumetric OCT images.
• To enhance endoscopic guidance during percutaneous nephrostomy procedures.
• To compare the performance of the 3D-CNN against 2D-CNNs and other state-of-the-art volumetric models.

Main Methods:

• A 3D-CNN was designed for volumetric OCT image analysis.
• Performance was assessed using 10-fold nested cross-validation on porcine kidney datasets.
• The 3D-CNN was benchmarked against 3D-ViT, 3D-DenseNet121, and M3T models.

Main Results:

• The 3D-CNN achieved an average test accuracy of 90.57%, outperforming 2D-CNN models (85.63%-88.22%).
• Inference latency for the 3D-CNN was 33 ms, significantly lower than competing volumetric architectures.
• Comparable inferencing accuracy was observed across the evaluated state-of-the-art models.

Conclusions:

• The developed 3D-CNN is a highly accurate and efficient tool for OCT image analysis.
• Its low inference latency makes it suitable for real-time applications in computer-aided diagnosis.
• This technology holds significant potential for improving OCT-guided surgical interventions.