AI-Assisted OCT Imaging for Core Needle Biopsy Guidance: The 1st in Humans Study

Area of Science:

• Medical Imaging
• Artificial Intelligence in Medicine
• Oncology

Background:

• Percutaneous image-guided biopsy success is limited by tissue heterogeneity (fat, necrosis, fibrosis).
• Current imaging (ultrasound, CT) lacks resolution to assess tissue composition for optimal biopsy site selection.
• Inadequate samples lead to repeat biopsies, increasing healthcare costs.

Purpose of the Study:

• To introduce a combined optical imaging/artificial intelligence (OI/AI) methodology for real-time tissue morphology assessment at the biopsy needle tip.
• To reduce the rate of non-diagnostic biopsy cores (currently up to 40%) due to low tumor content or high adipose tissue.
• To provide clinicians with real-time data for improved biopsy site selection.

Main Methods:

• Utilized micron-scale optical coherence tomography (OCT) imaging via a minimally invasive needle probe for detailed tissue structure analysis.
• Developed and employed a convolutional neural network (CNN)-driven AI software (U-net architecture) for automated segmentation of tumor regions from OCT scans.
• Conducted a clinical study with liver cancer patients, using expert-annotated OCT images to train and validate the AI algorithm.

Main Results:

• OCT imaging provided high-quality tissue images with ~10 mm axial and 20 mm lateral resolution.
• The U-Net AI model achieved an Area Under the Curve (AUC) of ~0.877 on the validation dataset.
• The AI model demonstrated approximately 90% agreement with human expert interpretations, indicating high accuracy.

Conclusions:

• A novel OCT instrument combined with AI software effectively assesses tissue composition at the biopsy needle tip.
• The system provides micron-scale resolution imaging, enabling AI-driven, real-time tissue type discrimination.
• This technology shows significant clinical potential for enhancing biopsy accuracy and success rates.