Classification of Pancreatic Cancer and Normal Tissue in 2D and 3D Optical Coherence Tomography Images Using...

Area of Science:

• Medical imaging
• Artificial intelligence in medicine
• Surgical oncology

Background:

• Optimal outcomes in pancreatic cancer depend on early and complete surgical resection (R0).
• Intraoperative guidance systems are needed to improve resection accuracy and reduce reliance on frozen sections.
• Optical coherence tomography (OCT) offers high-resolution imaging, and artificial intelligence (AI) can analyze complex image data.

Purpose of the Study:

• To evaluate the efficacy of convolutional neural networks (CNNs) in differentiating pancreatic ductal adenocarcinoma (PDAC) from normal pancreatic tissue using ex vivo OCT images.
• To assess the potential of AI-assisted OCT as a tool for real-time intraoperative guidance in pancreatic cancer surgery.

Main Methods:

• OCT scans were acquired from resected pancreatic specimens of 27 patients between October 2020 and April 2021.
• A 1310 nm OCT system was used to image tumor and adjacent normal tissues, followed by histopathological confirmation.
• Cross-validated CNN models (ResNet50, DenseNet121, MobileNetV2) were employed to analyze 2D and 3D OCT image inputs.

Main Results:

• The 3D DenseNet121 model achieved the highest performance, with an F1-score of 0.74, 72% sensitivity, and 81% specificity.
• Analysis utilized five-fold stratified cross-validation on 9040 2D and 3000 3D samples at 224 px resolution.
• Comparable results were observed with other CNN architectures, indicating robust performance.

Conclusions:

• AI-assisted OCT demonstrated accuracy in differentiating PDAC from normal pancreatic tissue ex vivo.
• This technology holds potential as a rapid intraoperative diagnostic adjunct for pancreatic cancer surgery.
• Further in vivo studies are necessary to validate performance and assess utility in margin evaluation.