Classification of esophageal cancer by using hyperspectral data

Area of Science:

• Oncology
• Medical Imaging
• Artificial Intelligence

Background:

• Esophageal cancer presents a significant global health challenge, particularly in Asia.
• Early diagnosis and accurate assessment of tumor extent are crucial for improving patient survival rates.
• Intraoperative techniques are vital for evaluating tumor margins during and after surgery.

Purpose of the Study:

• To investigate the utility of hyperspectral imaging (HSI) in combination with artificial intelligence for detecting cancerous esophageal tissue.
• To analyze differences in physiological parameters like water and hemoglobin content between healthy and cancerous esophageal tissues.
• To evaluate the performance of convolutional neural networks (CNNs) in classifying esophageal, stomach, and cancerous tissues.

Main Methods:

• Clinical study involving hyperspectral intraluminal recordings of esophageal and stomach tissue specimens.
• Application of two distinct convolutional neural networks (CNNs) for tissue classification.
• Analysis of physiological parameters including water and hemoglobin concentration.

Main Results:

• Significant differences in hemoglobin concentration and water content were observed between healthy and cancerous tissues, varying with tumor stage.
• A hybrid CNN achieved an average area under the curve (AUC) of 81% ± 3%, with 74% ± 8% sensitivity and 89% ± 2% specificity across all tissue types.
• The study demonstrated the potential of HSI in differentiating tissue types based on their spectral characteristics.

Conclusions:

• Hyperspectral imaging (HSI) shows promise as an intraoperative tool to support the detection of cancerous tissue in esophageal cancer.
• Further research involving detailed histopathological correlation is necessary to validate these findings.
• Future multicenter studies and data augmentation are recommended to enhance the reliability and generalizability of the results.