Diseased thyroid tissue classification in OCT images using deep learning: Towards surgical decision support.

Area of Science:

• Medical imaging
• Artificial intelligence in surgery
• Pathology diagnostics

Background:

• Optical coherence tomography (OCT) provides high-resolution intraoperative imaging for thyroid surgery.
• Interpreting OCT images for diseased tissue identification can be challenging for surgeons.
• Real-time automated analysis of OCT data is needed to support clinical decision-making.

Purpose of the Study:

• To investigate deep learning models for automated thyroid disease classification using OCT data.
• To evaluate the performance of 2D and 3D deep learning models on ex vivo thyroid tissue.
• To assess the utility of custom deep learning models on open-access datasets.

Main Methods:

• Collected 2D and 3D OCT data from ex vivo thyroid specimens of 22 patients.
• Trained and evaluated several deep learning models, including a 3D vision transformer.
• Validated custom models on two independent open-access datasets.

Main Results:

• The 3D vision transformer model achieved the highest performance on the thyroid dataset (MCC=0.79, accuracy=0.90) for normal vs. abnormal classification.
• Custom deep learning models demonstrated excellent performance on open-access datasets (MCC > 0.88, accuracy > 0.96).
• The models effectively classified normal versus abnormal thyroid tissue.

Conclusions:

• OCT combined with deep learning analysis shows promise for real-time, automated diseased tissue identification in thyroid surgery.
• Automated OCT analysis can significantly aid surgeons in intraoperative decision-making.
• This approach could enhance surgical precision and patient outcomes in thyroid procedures.