Deep convolutional neural networks for classifying head and neck cancer using hyperspectral imaging

Area of Science:

• Oncology
• Medical Imaging
• Artificial Intelligence

Background:

• Accurate margin diagnosis is crucial for successful cancer resection and patient remission.
• Hyperspectral imaging (HSI) offers a noncontact method to analyze tissue spectral and optical properties.
• Current methods for margin assessment can be time-consuming, impacting surgical workflow.

Purpose of the Study:

• To develop and evaluate a deep learning model for automated tissue classification using HSI.
• To assess the potential of HSI and convolutional neural networks (CNNs) for intraoperative margin assessment in head and neck cancer surgery.

Main Methods:

• Excised head and neck tissue samples (squamous-cell carcinoma and normal) were analyzed using HSI.
• A convolutional neural network (CNN) classifier was trained to differentiate between cancerous and normal tissue.
• CNN classifications were validated against manual annotations by a specialized head and neck pathologist.

Main Results:

• The CNN classifier demonstrated accurate classification of tissue samples based on HSI data.
• Preliminary results from 50 patients show promising performance of the HSI-CNN approach.
• The system has the potential for automatic tissue-labeling of surgical specimens.

Conclusions:

• Hyperspectral imaging integrated with deep learning shows significant potential for real-time cancer margin assessment.
• This technology could enhance surgical precision and improve patient outcomes in head and neck cancer.
• Further validation in larger patient cohorts is warranted to confirm clinical utility.