Precision Imaging for Early Detection of Esophageal Cancer

Area of Science:

• Medical imaging
• Artificial intelligence
• Oncology

Background:

• Early detection of early-stage esophageal cancer (ECA) is critical for effective treatment.
• Hyperspectral imaging (HSI) and artificial intelligence (AI) show potential for improving diagnostic accuracy.

Purpose of the Study:

• To evaluate the effectiveness of HSI and AI in detecting early-stage esophageal cancer.
• To compare the performance of AI models trained with HSI versus traditional narrow-band imaging (NBI).

Main Methods:

• Utilized a dataset of 3984 white light images (WLIs) and 3666 narrow-band images (NBIs).
• Employed the Yolov5 object detection algorithm to train four distinct AI models.
• Trained models using RGB-WLIs and NBIs, comparing HSI-enhanced data against standard NBI.

Main Results:

• HSI integration enhanced AI model prediction performance for early ECA detection.
• The HSI training model achieved an 8% increase in accuracy and 5-8% improvement in precision and recall.
• Models trained with WLIs demonstrated the most substantial performance gains.

Conclusions:

• Integrating HSI with AI technologies significantly boosts prediction performance for early ECA detection.
• Deep learning identification models show promise for medical detection research.
• AI-assisted HSI could enhance endoscopic diagnostic systems for faster, more accurate cancer detection.