Assisting Barrett's esophagus identification using endoscopic data augmentation based on Generative Adversarial...

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Oncology

Background:

• Barrett's esophagus cases are increasing, necessitating efficient diagnostic tools.
• Traditional diagnosis methods are time-consuming and resource-intensive.
• Computer-aided diagnosis shows promise but is limited by data scarcity.

Purpose of the Study:

• To develop a computer-aided approach using Generative Adversarial Networks (GANs) for enhanced Barrett's esophagus and adenocarcinoma detection.
• To address data limitations in machine learning for medical image analysis.
• To improve the accuracy and efficiency of early disease detection.

Main Methods:

• Utilized Deep Convolutional Generative Adversarial Networks (DCGANs) for data augmentation of endoscopic images.
• Employed Convolutional Neural Networks (CNNs), specifically LeNet-5 and AlexNet, for feature extraction and classification.
• Validated the methodology on two endoscopic image datasets, evaluating both full images and patch-split images.

Main Results:

• Achieved 90% accuracy for the patch-based approach and 85% for the image-based approach, using augmented datasets.
• Demonstrated statistically significant improvements compared to results from original, non-augmented datasets.
• Showcased that data augmentation with synthetic images outperformed original datasets and other recent methods.

Conclusions:

• Generative Adversarial Networks are effective for augmenting medical image datasets, crucial for accurate classification.
• The proposed computer-assisted approach significantly enhances the detection of Barrett's esophagus and adenocarcinoma.
• High-quality, augmented data is vital for advancing computer-assisted diagnosis in gastroenterology.