Classification of hyperspectral endocrine tissue images using support vector machines.

Area of Science:

• Surgical Innovation
• Medical Imaging Technology
• Computational Pathology

Background:

• Thyroidectomy is a common surgery with complex neck anatomy.
• Accurate tissue discrimination is crucial for surgical success.
• Hyperspectral imaging (HSI) offers a noninvasive, contactless solution for intraoperative guidance.

Purpose of the Study:

• To develop and evaluate a hyperspectral imaging (HSI) based method for real-time tissue discrimination during thyroidectomy.
• To assist surgeons in identifying key anatomical structures and materials.

Main Methods:

• Implementation of a supervised classification method for HSI data interpretation.
• Automatic discrimination of parathyroid, thyroid, recurrent laryngeal nerve, surrounding tissues, and instruments.
• Utilized support vector machine (SVM) for classification.
• Performed leave-one-patient-out cross-validation.

Main Results:

• Support vector machine (SVM) achieved classification and visualization in under 1.4 seconds.
• Mean patient accuracy of 68% ± 23% across all identified tissues and materials.
• The best performance was achieved using SVM.

Conclusions:

• The developed HSI method demonstrates potential for intraoperative surgical guidance.
• Further validation on a larger patient cohort is necessary to confirm these promising results.
• HSI can aid in distinguishing critical structures during thyroidectomy.