An Ensemble Learning Method for Detection of Head and Neck Squamous Cell Carcinoma Using Polarized Hyperspectral...

Area of Science:

• Medical Imaging
• Computational Pathology
• Oncology

Background:

• Head and neck squamous cell carcinoma (HNSCC) presents a significant mortality challenge.
• Accurate and early detection of HNSCC is crucial for effective treatment and improved patient outcomes.

Purpose of the Study:

• To develop a novel polarized hyperspectral imaging (PHSI) system for analyzing H&E-stained HNSCC pathological slides.
• To create a deep learning classification model utilizing convolutional neural networks (CNNs) for HNSCC detection.

Main Methods:

• Collected Stokes parameter hypercubes (S0, S1, S2, S3) from 56 HNSCC patients using a PHSI microscope.
• Synthesized pseudo-RGB images and applied data augmentation (rotations, flipping) to image patches.
• Developed a four-branch CNN architecture trained on individual Stokes parameters, followed by fine-tuning for final predictions.

Main Results:

• The PHSI system and CNN model achieved high accuracy, sensitivity, and specificity in classifying HNSCC.
• The developed model demonstrated robust performance on the constructed dataset.
• The study validates the potential of PHSI combined with deep learning for HNSCC diagnosis.

Conclusions:

• The Stokes-vector-derived PHSI system offers a promising approach for HNSCC detection in histopathology.
• Future research should focus on larger, more diverse datasets and advanced CNN architectures for enhanced classification, including tumor grading.