Classification of Benign-Malignant Thyroid Nodules Based on Hyperspectral Technology.

Area of Science:

• Medical Imaging
• Biomedical Engineering
• Cancer Diagnostics

Background:

• Thyroid cancer incidence is rising, necessitating improved diagnostic methods.
• Current intraoperative diagnosis of thyroid nodules is inefficient.
• Existing hyperspectral image classification often relies on pixel-based segmentation.

Purpose of the Study:

• To develop a rapid, hyperspectral-based method for diagnosing benign and malignant thyroid nodules during surgery.
• To propose a novel nodule classification approach using hyperspectral data blocks.
• To enhance the accuracy and efficiency of thyroid cancer diagnosis in clinical settings.

Main Methods:

• Acquisition of diverse thyroid nodule hyperspectral data using a custom system.
• Development of a V3Dnet algorithm, based on 3D CNN and VGG16, for classifying hyperspectral data blocks.
• Investigation of data block size impact on classification performance.

Main Results:

• A comprehensive classification model integrating data acquisition, preprocessing, and the V3Dnet algorithm was constructed.
• The V3Dnet algorithm achieved 84.63% classification accuracy for benign and malignant thyroid nodules with a 50 × 50 × 196 data block size.
• The study demonstrated the effectiveness of block-based hyperspectral data classification over pixel-based methods.

Conclusions:

• The proposed hyperspectral imaging method offers a rapid and accurate approach for intraoperative thyroid nodule diagnosis.
• The V3Dnet algorithm provides a robust tool for classifying thyroid nodules based on hyperspectral data blocks.
• This technology has the potential to improve surgical accuracy and support thyroid cancer research.